Upload folder using huggingface_hub

.gitattributes

@@ -58,3 +58,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+data/maze/100/train_A1_RWs_500K.txt filter=lfs diff=lfs merge=lfs -text
+data/maze/100/train_C1_RWs_10M.txt filter=lfs diff=lfs merge=lfs -text
+data/maze/100/train_C1_RWs_500K.txt filter=lfs diff=lfs merge=lfs -text
+data/maze/100/train_E1_RWs_500K.txt filter=lfs diff=lfs merge=lfs -text
+data/maze/100/train_H1_RWs_10M.txt filter=lfs diff=lfs merge=lfs -text
+data/maze/100/train_I1_RWs_10M.txt filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,31 @@
+# Output directories
+out/
+out2/
+# Keep result/plot dirs (small)
+!out/
+out/*
+!out/maze_kdetour/
+!out/plot/
+
+# Large training data
+data/maze/36/
+data/maze/100/
+
+# Python cache
+__pycache__/
+*.pyc
+*.pyo
+*.pyd
+.Python
+
+# Virtual environments
+venv/
+env/
+ENV/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+*~

analyze_maze.py

@@ -0,0 +1,407 @@
+import os
+import argparse
+import math
+from cli_utils import parse_count, format_count
+
+
+def analyze_maze_predictions(file_path, multitasks=False, no_task_tag=False):
+    """
+    Analyze maze predictions from test output file.
+
+    Args:
+        file_path: Path to the prediction file
+        multitasks: If True, separate analysis for Task A and Task B
+        no_task_tag: If True, data files do not contain task identifiers
+
+    Returns:
+        Dictionary with overall stats and per-task stats if multitasks=True
+    """
+    total = 0
+    correct = 0
+    illegal_direction = 0
+    incorrect_target = 0
+    syntax_error = 0
+    overall_high_conf = 0
+    overall_low_conf = 0
+
+    # Per-task statistics
+    taskA_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'illegal_direction': 0, 'incorrect_target': 0,
+                   'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskB_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'incorrect_target_label': 0, 'incorrect_neighbor_label': 0,
+                   'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskC_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'illegal_direction': 0, 'incorrect_target': 0,
+                   'incorrect_label': 0, 'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskD_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'illegal_direction': 0, 'incorrect_target': 0,
+                   'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskE_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'illegal_direction': 0, 'incorrect_target': 0,
+                   'incorrect_label': 0, 'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskF_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'incorrect_target_label': 0,
+                   'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskG_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'illegal_direction': 0, 'incorrect_target': 0,
+                   'high_conf_mistake': 0, 'low_conf_mistake': 0}
+    taskH_stats = {'total': 0, 'correct': 0, 'syntax_error': 0, 'illegal_direction': 0, 'incorrect_target': 0,
+                   'high_conf_mistake': 0, 'low_conf_mistake': 0}
+
+    task_stats_map = {
+        'A': taskA_stats, 'B': taskB_stats, 'C': taskC_stats,
+        'D': taskD_stats, 'E': taskE_stats, 'F': taskF_stats, 'G': taskG_stats,
+        'H': taskH_stats
+    }
+
+    with open(file_path, 'r') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            total += 1
+
+            parts = line.split()
+
+            # Detect task ID (A, B, C, D, E, F, G)
+            task_id = None
+            task_offset = 0
+
+            if not no_task_tag:
+                # Original logic with task tags
+                if len(parts) > 0 and parts[0] in ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H']:
+                    task_id = parts[0]
+                    task_offset = 1
+            else:
+                # In no_task_tag mode, read task type from parentheses in the line
+                if line.startswith('(') and ')' in line:
+                    end_paren = line.find(')')
+                    task_id_str = line[1:end_paren]
+                    if task_id_str in ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H']:
+                        task_id = task_id_str
+                        line_without_label = line[end_paren+1:].strip()
+                        parts = line_without_label.split()
+                        task_offset = 0
+
+            if task_id is None:
+                if ':' in line:
+                    colon_idx = line.index(':')
+                    prompt_part = line[:colon_idx].strip()
+                    prompt_tokens = prompt_part.split()
+
+                    if len(prompt_tokens) >= 2:
+                        answer_part = line[colon_idx + 1:].strip()
+                        answer_tokens = answer_part.split()
+                        if len(answer_tokens) >= 2 and answer_tokens[1] in ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j']:
+                            task_id = 'E'
+                        elif any(tok in ['L', 'R', 'F', 'T'] for tok in answer_tokens):
+                            task_id = 'C'
+                        elif len(prompt_tokens) == 2 and prompt_tokens[0].isdigit() and prompt_tokens[1].isdigit():
+                            task_id = 'A'
+                        elif prompt_tokens[0].isdigit() and prompt_tokens[1] in ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j']:
+                            task_id = 'D'
+                        elif prompt_tokens[0] in ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j']:
+                            task_id = 'F'
+                        elif len(prompt_tokens) == 4 and all(token.isdigit() for token in prompt_tokens):
+                            task_id = 'G'
+                        elif prompt_tokens[0].isdigit():
+                            if len(answer_tokens) == 5:
+                                task_id = 'B'
+
+            if multitasks and task_id in task_stats_map:
+                task_stats_map[task_id]['total'] += 1
+
+            # Confidence tracking
+            is_high = 'HIGH-CONF' in line
+            is_low = 'LOW-CONF' in line
+            if is_high:
+                overall_high_conf += 1
+                if task_id in task_stats_map:
+                    task_stats_map[task_id]['high_conf_mistake'] += 1
+            elif is_low:
+                overall_low_conf += 1
+                if task_id in task_stats_map:
+                    task_stats_map[task_id]['low_conf_mistake'] += 1
+
+            if 'is illegal' in line or 'exceeds feasible count' in line or 'invalid index' in line or 'no feasible edges' in line:
+                illegal_direction += 1
+                if task_id in task_stats_map and 'illegal_direction' in task_stats_map[task_id]:
+                    task_stats_map[task_id]['illegal_direction'] += 1
+                continue
+
+            if 'syntax error' in line:
+                syntax_error += 1
+                if task_id in task_stats_map:
+                    task_stats_map[task_id]['syntax_error'] += 1
+                continue
+
+            if 'incorrect neighbor label' in line:
+                if task_id == 'B':
+                    taskB_stats['incorrect_neighbor_label'] += 1
+                incorrect_target += 1
+                continue
+
+            if 'incorrect target node label' in line:
+                if task_id == 'B':
+                    taskB_stats['incorrect_target_label'] += 1
+                incorrect_target += 1
+                continue
+
+            if 'incorrect target label' in line:
+                if task_id == 'F':
+                    taskF_stats['incorrect_target_label'] += 1
+                incorrect_target += 1
+                continue
+
+            if 'incorrect label' in line:
+                if task_id == 'C':
+                    taskC_stats['incorrect_label'] += 1
+                elif task_id == 'E':
+                    taskE_stats['incorrect_label'] += 1
+                incorrect_target += 1
+                continue
+
+            if 'incorrect target node' in line:
+                incorrect_target += 1
+                if task_id in task_stats_map and 'incorrect_target' in task_stats_map[task_id]:
+                    task_stats_map[task_id]['incorrect_target'] += 1
+                continue
+
+            correct += 1
+            if task_id in task_stats_map:
+                task_stats_map[task_id]['correct'] += 1
+
+    stats = {
+        'total': total,
+        'correct': correct,
+        'syntax_error': syntax_error,
+        'illegal_direction': illegal_direction,
+        'incorrect_target': incorrect_target,
+        'high_conf_mistake': overall_high_conf,
+        'low_conf_mistake': overall_low_conf
+    }
+
+    if multitasks:
+        stats['taskA'] = taskA_stats
+        stats['taskB'] = taskB_stats
+        stats['taskC'] = taskC_stats
+        stats['taskD'] = taskD_stats
+        stats['taskE'] = taskE_stats
+        stats['taskF'] = taskF_stats
+        stats['taskG'] = taskG_stats
+        stats['taskH'] = taskH_stats
+
+    return stats
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Analyze prediction results from test_maze.py')
+    parser.add_argument('--ckpt_iter', type=int, default=10000, help='Checkpoint iteration')
+    parser.add_argument('--model', type=str, default='transformer', choices=['transformer', 'transformer-rope', 'transformer-nextlat', 'mamba', 'mamba2', 'gated-deltanet', 'gru'],
+                        help='Model architecture; selects the out/<model>/ directory')
+    parser.add_argument('--config', type=str, default='1_1_120', help='Model config')
+    parser.add_argument('--dataset', type=str, default='maze', help='Dataset name')
+    parser.add_argument('--num_nodes', type=int, default=100, help='Number of nodes')
+    parser.add_argument('--num_of_paths', type=int, default=20, help='Number of paths')
+    parser.add_argument('--multitasks', action=argparse.BooleanOptionalAction, default=True,
+                        help='Use multitask data (default: True)')
+    parser.add_argument('--num_train_dataset', type=parse_count, default=50000,
+                        help='Number of multitask training entries (supports K/M/B, default: 50000)')
+    parser.add_argument('--num_test_dataset', type=parse_count, default=10000,
+                        help='Number of multitask test entries (supports K/M/B, default: 10000)')
+    parser.add_argument('--tasks', type=str, default='A1',
+                        help='Task specification (e.g., A1, A1B1, A3B2, A1D1F1). Default: A1')
+    parser.add_argument('--CL', action=argparse.BooleanOptionalAction, default=False,
+                        help='Task C turn-label mode (default: False)')
+    parser.add_argument('--batch_size', type=int, default=100,
+                        help='Batch size used during prediction (matches test_maze.py)')
+    parser.add_argument('--num_iters', type=int, default=10,
+                        help='Number of batches used during prediction (matches test_maze.py)')
+    parser.add_argument('--path_type', type=str, default='RWa', choices=['RWc', 'RWa', 'RWs'],
+                        help='Path generation type: RWc (random walk with cycles), RWa (random walk acyclic, default), RWs (single source random walk).')
+    parser.add_argument('--partial', action='store_true', default=False,
+                        help='Analyze partial prefix test results (default: False)')
+    parser.add_argument('--temperature', type=float, default=1.0,
+                        help='Sampling temperature used during prediction (default: 1.0). Affects output filenames.')
+    # Add --no_task_tag argument
+    parser.add_argument('--no_task_tag', action='store_true', default=False,
+                        help='Data files do not contain task identifiers (A, B, C, etc.). This should match the setting used during data generation and testing.')
+    parser.add_argument('--PostGRU', action='store_true', default=False,
+                        help='Analyze PostGRU predictions (adds _PGR suffix to filenames)')
+    parser.add_argument('--NLS', action='store_true', default=False,
+                        help='Analyze NLS predictions (adds _NLS suffix to filenames)')
+    args = parser.parse_args()
+
+    tasks_str = args.tasks
+    tasks_tag = f"{tasks_str}_CL" if args.CL else tasks_str
+    # Add path type tag for filenames (RWc = cyclic, RWa = acyclic, RWs = single source)
+    path_type_tag = args.path_type
+    tasks_tag = f"{tasks_tag}_{path_type_tag}"
+    # Add _NT_ tag to tasks_tag when no_task_tag is enabled
+    if args.no_task_tag:
+        tasks_tag = f"{tasks_tag}_NT"
+    # Add _NL tag for transformer-nextlat
+    if args.model == 'transformer-nextlat':
+        tasks_tag = f"{tasks_tag}_NL"
+    # Add _PGR tag when PostGRU is enabled
+    if args.PostGRU:
+        tasks_tag = f"{tasks_tag}_PGR"
+    # Add _NLS tag when NLS is enabled
+    if args.NLS:
+        tasks_tag = f"{tasks_tag}_NLS"
+    test_dataset_label = format_count(args.num_test_dataset)
+    run_test_label = args.batch_size * args.num_iters
+    nt_suffix = '_NT' if args.no_task_tag else ''
+    out_dir = f'out/{args.model.replace("-", "_")}/{args.dataset}_{args.config}_{args.num_nodes}{nt_suffix}/'
+
+
+    def pick_first_existing(paths):
+        for path in paths:
+            if os.path.exists(path):
+                return path
+        return paths[0]
+
+
+    # Add _partial suffix if partial mode is enabled
+    partial_suffix = '_partial' if args.partial else ''
+    # Add temperature suffix when temperature is not default (1.0)
+    temp_suffix = f'_t{args.temperature}' if args.temperature != 1.0 else ''
+
+    pred_candidates = (
+        [
+            os.path.join(out_dir, f'pred_test_{tasks_tag}_{args.ckpt_iter}_{run_test_label}{temp_suffix}{partial_suffix}.txt'),
+            # primary (matches test_maze.py)
+            os.path.join(out_dir, f'pred_test_{tasks_tag}_{args.ckpt_iter}_{test_dataset_label}{temp_suffix}{partial_suffix}.txt'),
+            os.path.join(out_dir,
+                         f'pred_test_{tasks_tag}_{args.ckpt_iter}_{args.num_test_dataset}{temp_suffix}{partial_suffix}.txt'),
+            os.path.join(out_dir, f'pred_test_{tasks_str}_{args.ckpt_iter}_{run_test_label}{temp_suffix}{partial_suffix}.txt'),
+            os.path.join(out_dir, f'pred_test_{tasks_str}_{args.ckpt_iter}_{test_dataset_label}{temp_suffix}{partial_suffix}.txt'),
+            os.path.join(out_dir,
+                         f'pred_test_{tasks_str}_{args.ckpt_iter}_{args.num_test_dataset}{temp_suffix}{partial_suffix}.txt'),
+        ]
+        if args.multitasks else [
+            os.path.join(out_dir, f'pred_test_{args.ckpt_iter}_{args.num_of_paths}{temp_suffix}{partial_suffix}.txt')]
+    )
+    file_path = pick_first_existing(pred_candidates)
+
+    if os.path.exists(file_path):
+        # Ensure output directory exists
+        os.makedirs(out_dir, exist_ok=True)
+
+        # Analyze predictions
+        stats = analyze_maze_predictions(file_path, multitasks=args.multitasks, no_task_tag=args.no_task_tag)
+
+
+        def pct_and_se(count, total):
+            if total <= 0:
+                return 0.0, 0.0
+            p = count / total
+            se = math.sqrt(p * (1 - p) / total) * 100
+            return p * 100, se
+
+
+        # Format output
+        separator = "=" * 70
+        output_lines = [
+            separator,
+            "Accuracy Test Results",
+            separator,
+            f"Task tag: {'DISABLED' if args.no_task_tag else 'ENABLED'}",
+            f"Config: {args.config}",
+            f"Checkpoint iteration: {args.ckpt_iter}",
+            f"Number of nodes: {args.num_nodes}",
+            f"Task configuration: {args.tasks}" if args.multitasks else "",
+            separator,
+        ]
+
+        # Overall statistics
+        total_preds = stats['total'] if stats['total'] > 0 else 1
+        corr_pct, corr_se = pct_and_se(stats['correct'], total_preds)
+        syn_pct, _ = pct_and_se(stats['syntax_error'], total_preds)
+        ill_pct, _ = pct_and_se(stats['illegal_direction'], total_preds)
+        tgt_pct, _ = pct_and_se(stats['incorrect_target'], total_preds)
+        high_pct, _ = pct_and_se(stats['high_conf_mistake'], total_preds)
+        low_pct, _ = pct_and_se(stats['low_conf_mistake'], total_preds)
+
+        output_lines.extend([
+            "OVERALL STATISTICS:",
+            f"  Total predictions: {stats['total']}",
+            f"  Correct (accuracy with standard error): {stats['correct']} ({corr_pct:.2f}% ± {corr_se:.2f}%)",
+            f"  Syntax error: {stats['syntax_error']} ({syn_pct:.2f}%)",
+            f"  Illegal direction: {stats['illegal_direction']} ({ill_pct:.2f}%)",
+            f"  Incorrect target: {stats['incorrect_target']} ({tgt_pct:.2f}%)",
+            f"    - High confidence mistakes: {stats['high_conf_mistake']} ({high_pct:.2f}%)",
+            f"    - Low confidence mistakes: {stats['low_conf_mistake']} ({low_pct:.2f}%)",
+        ])
+
+        # Per-task statistics if multitasks
+        if args.multitasks:
+            task_mapping = {
+                'taskA': ('A', 'Pathfinding'),
+                'taskB': ('B', 'Target Identification'),
+                'taskC': ('C', 'Turn-based pathfinding'),
+                'taskD': ('D', 'Pathfinding to label'),
+                'taskE': ('E', 'Pathfinding with labels'),
+                'taskF': ('F', 'Target label identification'),
+                'taskG': ('G', 'Reachability choice'),
+                'taskH': ('H', 'Relative clockwise-index path')
+            }
+
+            for key, (tid, name) in task_mapping.items():
+                if key in stats and stats[key]['total'] > 0:
+                    s = stats[key]
+                    t_total = s['total']
+                    t_corr, t_se = pct_and_se(s['correct'], t_total)
+                    t_syn, _ = pct_and_se(s['syntax_error'], t_total)
+                    t_high, _ = pct_and_se(s['high_conf_mistake'], t_total)
+                    t_low, _ = pct_and_se(s['low_conf_mistake'], t_total)
+
+                    output_lines.extend([
+                        "",
+                        separator,
+                        f"TASK {tid} ({name}) STATISTICS:",
+                        f"  Total: {t_total}",
+                        f"  Correct (accuracy with standard error): {s['correct']} ({t_corr:.2f}% ± {t_se:.2f}%)",
+                        f"  Syntax error: {s['syntax_error']} ({t_syn:.2f}%)",
+                    ])
+
+                    if 'illegal_direction' in s:
+                        t_ill, _ = pct_and_se(s['illegal_direction'], t_total)
+                        output_lines.append(f"  Illegal direction: {s['illegal_direction']} ({t_ill:.2f}%)")
+
+                    if tid == 'B':
+                        t_lbl, _ = pct_and_se(s['incorrect_target_label'], t_total)
+                        t_nbr, _ = pct_and_se(s['incorrect_neighbor_label'], t_total)
+                        output_lines.append(
+                            f"  Incorrect target node label: {s['incorrect_target_label']} ({t_lbl:.2f}%)")
+                        output_lines.append(
+                            f"  Incorrect neighbor label: {s['incorrect_neighbor_label']} ({t_nbr:.2f}%)")
+                    elif tid == 'F':
+                        t_lbl, _ = pct_and_se(s['incorrect_target_label'], t_total)
+                        output_lines.append(f"  Incorrect target label: {s['incorrect_target_label']} ({t_lbl:.2f}%)")
+                    else:
+                        if 'incorrect_target' in s:
+                            t_tgt, _ = pct_and_se(s['incorrect_target'], t_total)
+                            output_lines.append(f"  Incorrect target: {s['incorrect_target']} ({t_tgt:.2f}%)")
+
+                    if 'incorrect_label' in s:
+                        t_lbl, _ = pct_and_se(s['incorrect_label'], t_total)
+                        lbl_text = "Incorrect label (CL mode)" if tid == 'C' else "Incorrect label"
+                        output_lines.append(f"  {lbl_text}: {s['incorrect_label']} ({t_lbl:.2f}%)")
+
+                    output_lines.extend([
+                        f"    - High confidence mistakes: {s['high_conf_mistake']} ({t_high:.2f}%)",
+                        f"    - Low confidence mistakes: {s['low_conf_mistake']} ({t_low:.2f}%)",
+                    ])
+
+        output_lines.append(separator)
+        output_text = "\n".join(output_lines)
+
+        # Print to console
+        print("\n" + output_text + "\n")
+
+        # Save to file
+        output_file = os.path.join(
+            out_dir,
+            f"accuracy_{tasks_tag}_{args.ckpt_iter}_{args.num_test_dataset}{temp_suffix}{partial_suffix}.txt" if args.multitasks else f"accuracy_{args.ckpt_iter}_{args.num_of_paths}{temp_suffix}{partial_suffix}.txt"
+        )
+        with open(output_file, 'w') as f:
+            f.write(output_text + "\n")
+
+        print(f"Results saved to {output_file}")
+    else:
+        print(f"File {file_path} not found.")

analyze_simple.py

@@ -0,0 +1,56 @@
+import os
+import argparse
+
+def analyze_predictions(file_path):
+    total = 0
+    correct = 0
+    incorrect_start_end = 0
+    non_existence = 0
+    wrong_syntax = 0
+    
+    with open(file_path, 'r') as f:
+        for line in f:
+            line = line.strip()
+            if not line:
+                continue
+            total += 1
+            parts = line.split()
+            # Find where symbol starts (first non-digit, non-'x')
+            symbol_start = len(parts)
+            for i, p in enumerate(parts):
+                if not (p.isdigit() or p == 'x'):
+                    symbol_start = i
+                    break
+            symbol = ' '.join(parts[symbol_start:]).strip()
+            if symbol == '':
+                correct += 1
+            elif symbol == 'incorrect start/end':
+                incorrect_start_end += 1
+            elif symbol.startswith('non-existence path'):
+                non_existence += 1
+            elif symbol == 'wrong syntax':
+                wrong_syntax += 1
+            # else: other, but according to code, only these
+    
+    print(f"Total paths: {total}")
+    print(f"Correct paths: {correct} ({correct/total*100:.2f}%)")
+    print(f"Incorrect start/end: {incorrect_start_end} ({incorrect_start_end/total*100:.2f}%)")
+    print(f"Non-existence path: {non_existence} ({non_existence/total*100:.2f}%)")
+    print(f"Wrong syntax: {wrong_syntax} ({wrong_syntax/total*100:.2f}%)")
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Analyze prediction results from test_simple.py')
+    parser.add_argument('--ckpt_iter', type=int, default=10000, help='Checkpoint iteration')
+    parser.add_argument('--config', type=str, default='1_1_120', help='Model config')
+    parser.add_argument('--dataset', type=str, default='simple_graph', help='Dataset name')
+    parser.add_argument('--num_nodes', type=int, default=100, help='Number of nodes')
+    parser.add_argument('--num_of_paths', type=int, default=20, help='Number of paths')
+    args = parser.parse_args()
+    
+    out_dir = f'out/{args.dataset}_{args.config}_{args.num_nodes}/'
+    file_path = os.path.join(out_dir, f'pred_test_{args.ckpt_iter}.txt')
+    
+    if os.path.exists(file_path):
+        analyze_predictions(file_path)
+    else:
+        print(f"File {file_path} not found.")

cli_utils.py

@@ -0,0 +1,112 @@
+import argparse
+import random
+import re
+
+
+def parse_count(value):
+    """Parse integers with optional K/M/B suffixes (e.g., 50K, 3M)."""
+    if isinstance(value, int):
+        return value
+    if isinstance(value, str):
+        text = value.strip().replace('_', '').lower()
+        match = re.fullmatch(r'(\d+(?:\.\d+)?)([kmb]?)', text)
+        if not match:
+            raise argparse.ArgumentTypeError(f"Invalid count: {value}")
+        number, suffix = match.groups()
+        multipliers = {'': 1, 'k': 1_000, 'm': 1_000_000, 'b': 1_000_000_000}
+        return int(float(number) * multipliers[suffix])
+    raise argparse.ArgumentTypeError(f"Invalid count type: {type(value)}")
+
+
+def format_count(value):
+    """Format an integer count using K/M/B suffixes for filenames (e.g., 3000 -> 3K)."""
+    n = int(value)
+    for suffix, factor in [('B', 1_000_000_000), ('M', 1_000_000), ('K', 1_000)]:
+        if n >= factor:
+            if n % factor == 0:
+                return f"{n // factor}{suffix}"
+            short = n / factor
+            return f"{short:.1f}".rstrip('0').rstrip('.') + suffix
+    return str(n)
+
+
+def parse_task_distribution(tasks_str, default_task='A'):
+    """Parse task weights like 'A1C1' into a dict of task -> weight."""
+    if tasks_str is None or tasks_str == '':
+        return {default_task: 1}
+    matches = re.findall(r'([A-I])(\d+)', tasks_str)
+    if not matches:
+        raise ValueError(f"Invalid task specification: '{tasks_str}'. Expected format like 'A1' or 'A1C1'.")
+
+    weights = {}
+    for task, count_str in matches:
+        if task in weights:
+            raise ValueError(f"Duplicate task ID: {task}")
+        count = int(count_str)
+        if count <= 0:
+            raise ValueError(f"Task {task} ratio must be positive, got {count}")
+        weights[task] = count
+    return weights if weights else {default_task: 1}
+
+
+def sample_task(weights, allowed_tasks=None, default_task='A'):
+    """Sample a task ID from a weight dict, respecting an allowed set."""
+    allowed_set = set(allowed_tasks) if allowed_tasks else set(weights.keys())
+    filtered = {k: v for k, v in weights.items() if k in allowed_set}
+    if not filtered:
+        filtered = {default_task: weights.get(default_task, 1)}
+
+    total = sum(filtered.values())
+    r = random.uniform(0, total)
+    upto = 0
+    for task, weight in filtered.items():
+        upto += weight
+        if upto >= r:
+            return task
+    return next(iter(filtered))
+
+
+def directions_to_turns(direction_seq, start_orientation='E'):
+    """Convert absolute NESW directions into relative turns (L/R/F/T) assuming starting orientation east."""
+    orientation = start_orientation
+    turns = []
+    left_of = {'N': 'W', 'W': 'S', 'S': 'E', 'E': 'N'}
+    right_of = {v: k for k, v in left_of.items()}
+    opposite_of = {'N': 'S', 'S': 'N', 'E': 'W', 'W': 'E'}
+
+    for direction in direction_seq:
+        if direction == orientation:
+            turn = 'F'
+        elif left_of[orientation] == direction:
+            turn = 'L'
+        elif right_of[orientation] == direction:
+            turn = 'R'
+        else:
+            turn = 'T'
+        turns.append(turn)
+        orientation = direction
+
+    return turns
+
+
+def turns_to_directions(turn_seq, start_orientation='E'):
+    """Convert relative turns back to absolute NESW directions."""
+    orientation = start_orientation
+    directions = []
+    left_of = {'N': 'W', 'W': 'S', 'S': 'E', 'E': 'N'}
+    right_of = {v: k for k, v in left_of.items()}
+    opposite_of = {'N': 'S', 'S': 'N', 'E': 'W', 'W': 'E'}
+
+    for turn in turn_seq:
+        if turn == 'F':
+            new_orientation = orientation
+        elif turn == 'L':
+            new_orientation = left_of[orientation]
+        elif turn == 'R':
+            new_orientation = right_of[orientation]
+        else:  # 'T'
+            new_orientation = opposite_of[orientation]
+        directions.append(new_orientation)
+        orientation = new_orientation
+
+    return directions

data/maze/100/maze_A1_RWs_10_0.6.txt

@@ -0,0 +1,20 @@
++---+---+---+   +---+---+---+---+   +
+ c  |i  |d   c  |h  |h   e  |a  |f  |f  
++---+   +---+   +---+   +   +---+---+
+ c  |f  |g   f   b  |a   j   j  |d   f  
++   +   +---+---+---+   +---+---+---+
+|g  |h  |j   j  |d  |j   e  |c  |b  |i  
++---+   +---+   +   +---+---+   +---+
+ d  |d  |f   e   i   a  |j  |c   i   e  
++   +---+   +   +   +---+---+---+   +
+ b  |d  |j  |d   e   d   e  |d  |f   d  
++---+---+   +---+---+   +---+---+   +
+|g  |h  |j   e   f  |d  |b   d  |h   a  
++   +---+   +---+---+   +   +   +---+
+|j  |g   e  |f  |c  |j  |d  |b  |a  |f  
++---+---+---+---+---+---+---+---+   +
+ h   i   c  |d  |j   f  |e  |h   f  |j  
++   +   +---+   +   +---+---+---+---+
+ a  |d   e   g   i  |g  |f  |h  |d   a  
++---+---+   +   +---+   +---+   +   +
+ c   e   g   d   h   i   i   d   b   a  

data/maze/100/maze_C1_RWs_10_0.6.txt

@@ -0,0 +1,20 @@
++   +---+   +---+---+---+---+---+---+
+ b  |j   h   h   h   j   f   e  |f   j  
++---+---+---+---+---+---+   +   +---+
+ f   i  |a  |b  |e   a  |a   c  |c  |f  
++   +---+---+---+---+---+---+---+---+
+|i  |d  |c  |b   h  |g   b  |b   i  |a  
++---+---+---+   +   +---+---+   +---+
+|c   f   c   i   f  |h   f  |b   i   c  
++   +---+---+---+---+   +   +   +---+
+ a   a   j   e  |g  |b  |a  |a  |a  |i  
++---+---+---+   +---+---+   +---+---+
+|d   g  |d  |g  |i   e  |a   a  |e   d  
++   +---+---+---+---+---+   +   +---+
+ b   i   f   a   c  |f   i   d  |d   j  
++---+   +---+---+---+---+---+   +   +
+|b  |e  |j   a  |d   j  |g   g  |a  |d  
++---+---+---+---+---+---+   +   +---+
+|d  |i   f   d   h  |g  |i   c  |b  |f  
++   +   +   +---+---+---+---+   +---+
+ f   d   h   e   g   f   j   h   e   h  

data/maze/100/maze_E1_RWs_10_0.6.txt

@@ -0,0 +1,20 @@
++   +---+---+   +---+---+---+   +   +
+|i  |e  |c  |f  |a  |c   i   j   d   e  
++---+---+   +---+---+---+---+   +   +
+ c   g  |c  |j   d   j   i  |i   f   c  
++---+---+   +---+---+---+---+---+---+
+ j  |i  |b  |d  |j   e  |a   c  |h  |g  
++   +---+---+   +   +---+---+---+   +
+ d  |g   f   b   c  |f  |c  |i  |b   d  
++---+---+---+---+   +   +   +---+---+
+|i  |h   f  |h  |h  |h  |h  |h  |j  |b  
++---+---+   +   +---+---+---+---+---+
+ e  |e  |d  |e  |f  |d   j  |g  |b  |e  
++   +---+   +   +   +   +---+---+   +
+|f  |a   e  |e  |f   g   g  |h   a  |j  
++---+---+   +---+   +   +---+---+   +
+|h   d   j   f  |i   f   i   c  |j   i  
++---+---+---+   +---+   +   +   +---+
+|d  |b  |e  |e  |e  |f  |h  |g  |i  |f  
++   +---+   +   +---+---+---+---+---+
+ d   e   j   j   c   g   b   f   e   b  

data/maze/100/maze_H1_RWs_10_0.6.txt

@@ -0,0 +1,20 @@
++   +---+---+---+   +   +   +---+   +
+ e  |c   g   c  |h   b   h  |c   g  |f  
++---+---+---+---+---+   +---+   +   +
+|f   d   f  |i   c  |a  |b  |c  |d   e  
++   +---+---+---+---+   +---+---+---+
+|i   f  |b   j  |f   j  |c  |h  |g   b  
++---+   +---+---+---+---+   +   +---+
+|e   b   g  |i   b  |f  |j  |g  |b  |j  
++---+   +---+---+---+   +---+---+---+
+ a   c  |g  |e   e   d  |j   c  |b  |c  
++---+---+   +---+   +   +---+---+---+
+ i  |e   h  |b   a  |a   h  |j   a   f  
++   +   +   +---+   +---+   +   +---+
+|c   a   g   j   c   b   e  |e   d  |b  
++---+   +---+---+---+   +   +---+   +
+|b   f  |g   d  |g  |i   j  |b   e  |d  
++---+---+---+---+---+---+---+---+---+
+|e  |h  |e  |e   i   j  |b  |a  |i  |d  
++---+---+   +---+---+---+---+   +   +
+ h   c   e   d   b   g   j   b   f   j  

data/maze/100/maze_I1_RWs_10_0.6.txt

@@ -0,0 +1,20 @@
++---+---+---+   +---+---+---+---+   +
+ c  |i  |d   c  |h  |h   e  |a  |f  |f  
++---+   +---+   +---+   +   +---+---+
+ c  |f  |g   f   b  |a   j   j  |d   f  
++   +   +---+---+---+   +---+---+---+
+|g  |h  |j   j  |d  |j   e  |c  |b  |i  
++---+   +---+   +   +---+---+   +---+
+ d  |d  |f   e   i   a  |j  |c   i   e  
++   +---+   +   +   +---+---+---+   +
+ b  |d  |j  |d   e   d   e  |d  |f   d  
++---+---+   +---+---+   +---+---+   +
+|g  |h  |j   e   f  |d  |b   d  |h   a  
++   +---+   +---+---+   +   +   +---+
+|j  |g   e  |f  |c  |j  |d  |b  |a  |f  
++---+---+---+---+---+---+---+---+   +
+ h   i   c  |d  |j   f  |e  |h   f  |j  
++   +   +---+   +   +---+---+---+---+
+ a  |d   e   g   i  |g  |f  |h  |d   a  
++---+---+   +   +---+   +---+   +   +
+ c   e   g   d   h   i   i   d   b   a  

data/maze/100/maze_graph_A1_RWs.graphml

@@ -0,0 +1,520 @@
+<?xml version='1.0' encoding='utf-8'?>
+<graphml xmlns="http://graphml.graphdrawing.org/xmlns" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://graphml.graphdrawing.org/xmlns http://graphml.graphdrawing.org/xmlns/1.0/graphml.xsd">
+  <key id="d0" for="node" attr.name="label" attr.type="string" />
+  <graph edgedefault="directed">
+    <node id="0">
+      <data key="d0">c</data>
+    </node>
+    <node id="1">
+      <data key="d0">i</data>
+    </node>
+    <node id="2">
+      <data key="d0">d</data>
+    </node>
+    <node id="3">
+      <data key="d0">c</data>
+    </node>
+    <node id="4">
+      <data key="d0">h</data>
+    </node>
+    <node id="5">
+      <data key="d0">h</data>
+    </node>
+    <node id="6">
+      <data key="d0">e</data>
+    </node>
+    <node id="7">
+      <data key="d0">a</data>
+    </node>
+    <node id="8">
+      <data key="d0">f</data>
+    </node>
+    <node id="9">
+      <data key="d0">f</data>
+    </node>
+    <node id="10">
+      <data key="d0">c</data>
+    </node>
+    <node id="11">
+      <data key="d0">f</data>
+    </node>
+    <node id="12">
+      <data key="d0">g</data>
+    </node>
+    <node id="13">
+      <data key="d0">f</data>
+    </node>
+    <node id="14">
+      <data key="d0">b</data>
+    </node>
+    <node id="15">
+      <data key="d0">a</data>
+    </node>
+    <node id="16">
+      <data key="d0">j</data>
+    </node>
+    <node id="17">
+      <data key="d0">j</data>
+    </node>
+    <node id="18">
+      <data key="d0">d</data>
+    </node>
+    <node id="19">
+      <data key="d0">f</data>
+    </node>
+    <node id="20">
+      <data key="d0">g</data>
+    </node>
+    <node id="21">
+      <data key="d0">h</data>
+    </node>
+    <node id="22">
+      <data key="d0">j</data>
+    </node>
+    <node id="23">
+      <data key="d0">j</data>
+    </node>
+    <node id="24">
+      <data key="d0">d</data>
+    </node>
+    <node id="25">
+      <data key="d0">j</data>
+    </node>
+    <node id="26">
+      <data key="d0">e</data>
+    </node>
+    <node id="27">
+      <data key="d0">c</data>
+    </node>
+    <node id="28">
+      <data key="d0">b</data>
+    </node>
+    <node id="29">
+      <data key="d0">i</data>
+    </node>
+    <node id="30">
+      <data key="d0">d</data>
+    </node>
+    <node id="31">
+      <data key="d0">d</data>
+    </node>
+    <node id="32">
+      <data key="d0">f</data>
+    </node>
+    <node id="33">
+      <data key="d0">e</data>
+    </node>
+    <node id="34">
+      <data key="d0">i</data>
+    </node>
+    <node id="35">
+      <data key="d0">a</data>
+    </node>
+    <node id="36">
+      <data key="d0">j</data>
+    </node>
+    <node id="37">
+      <data key="d0">c</data>
+    </node>
+    <node id="38">
+      <data key="d0">i</data>
+    </node>
+    <node id="39">
+      <data key="d0">e</data>
+    </node>
+    <node id="40">
+      <data key="d0">b</data>
+    </node>
+    <node id="41">
+      <data key="d0">d</data>
+    </node>
+    <node id="42">
+      <data key="d0">j</data>
+    </node>
+    <node id="43">
+      <data key="d0">d</data>
+    </node>
+    <node id="44">
+      <data key="d0">e</data>
+    </node>
+    <node id="45">
+      <data key="d0">d</data>
+    </node>
+    <node id="46">
+      <data key="d0">e</data>
+    </node>
+    <node id="47">
+      <data key="d0">d</data>
+    </node>
+    <node id="48">
+      <data key="d0">f</data>
+    </node>
+    <node id="49">
+      <data key="d0">d</data>
+    </node>
+    <node id="50">
+      <data key="d0">g</data>
+    </node>
+    <node id="51">
+      <data key="d0">h</data>
+    </node>
+    <node id="52">
+      <data key="d0">j</data>
+    </node>
+    <node id="53">
+      <data key="d0">e</data>
+    </node>
+    <node id="54">
+      <data key="d0">f</data>
+    </node>
+    <node id="55">
+      <data key="d0">d</data>
+    </node>
+    <node id="56">
+      <data key="d0">b</data>
+    </node>
+    <node id="57">
+      <data key="d0">d</data>
+    </node>
+    <node id="58">
+      <data key="d0">h</data>
+    </node>
+    <node id="59">
+      <data key="d0">a</data>
+    </node>
+    <node id="60">
+      <data key="d0">j</data>
+    </node>
+    <node id="61">
+      <data key="d0">g</data>
+    </node>
+    <node id="62">
+      <data key="d0">e</data>
+    </node>
+    <node id="63">
+      <data key="d0">f</data>
+    </node>
+    <node id="64">
+      <data key="d0">c</data>
+    </node>
+    <node id="65">
+      <data key="d0">j</data>
+    </node>
+    <node id="66">
+      <data key="d0">d</data>
+    </node>
+    <node id="67">
+      <data key="d0">b</data>
+    </node>
+    <node id="68">
+      <data key="d0">a</data>
+    </node>
+    <node id="69">
+      <data key="d0">f</data>
+    </node>
+    <node id="70">
+      <data key="d0">h</data>
+    </node>
+    <node id="71">
+      <data key="d0">i</data>
+    </node>
+    <node id="72">
+      <data key="d0">c</data>
+    </node>
+    <node id="73">
+      <data key="d0">d</data>
+    </node>
+    <node id="74">
+      <data key="d0">j</data>
+    </node>
+    <node id="75">
+      <data key="d0">f</data>
+    </node>
+    <node id="76">
+      <data key="d0">e</data>
+    </node>
+    <node id="77">
+      <data key="d0">h</data>
+    </node>
+    <node id="78">
+      <data key="d0">f</data>
+    </node>
+    <node id="79">
+      <data key="d0">j</data>
+    </node>
+    <node id="80">
+      <data key="d0">a</data>
+    </node>
+    <node id="81">
+      <data key="d0">d</data>
+    </node>
+    <node id="82">
+      <data key="d0">e</data>
+    </node>
+    <node id="83">
+      <data key="d0">g</data>
+    </node>
+    <node id="84">
+      <data key="d0">i</data>
+    </node>
+    <node id="85">
+      <data key="d0">g</data>
+    </node>
+    <node id="86">
+      <data key="d0">f</data>
+    </node>
+    <node id="87">
+      <data key="d0">h</data>
+    </node>
+    <node id="88">
+      <data key="d0">d</data>
+    </node>
+    <node id="89">
+      <data key="d0">a</data>
+    </node>
+    <node id="90">
+      <data key="d0">c</data>
+    </node>
+    <node id="91">
+      <data key="d0">e</data>
+    </node>
+    <node id="92">
+      <data key="d0">g</data>
+    </node>
+    <node id="93">
+      <data key="d0">d</data>
+    </node>
+    <node id="94">
+      <data key="d0">h</data>
+    </node>
+    <node id="95">
+      <data key="d0">i</data>
+    </node>
+    <node id="96">
+      <data key="d0">i</data>
+    </node>
+    <node id="97">
+      <data key="d0">d</data>
+    </node>
+    <node id="98">
+      <data key="d0">b</data>
+    </node>
+    <node id="99">
+      <data key="d0">a</data>
+    </node>
+    <edge source="0" target="1" />
+    <edge source="1" target="0" />
+    <edge source="1" target="11" />
+    <edge source="1" target="2" />
+    <edge source="2" target="1" />
+    <edge source="2" target="12" />
+    <edge source="2" target="3" />
+    <edge source="3" target="2" />
+    <edge source="4" target="14" />
+    <edge source="4" target="5" />
+    <edge source="5" target="4" />
+    <edge source="5" target="15" />
+    <edge source="5" target="6" />
+    <edge source="6" target="5" />
+    <edge source="6" target="7" />
+    <edge source="7" target="6" />
+    <edge source="7" target="17" />
+    <edge source="7" target="8" />
+    <edge source="8" target="7" />
+    <edge source="8" target="18" />
+    <edge source="9" target="19" />
+    <edge source="10" target="11" />
+    <edge source="11" target="1" />
+    <edge source="11" target="10" />
+    <edge source="11" target="21" />
+    <edge source="12" target="2" />
+    <edge source="12" target="22" />
+    <edge source="12" target="13" />
+    <edge source="13" target="12" />
+    <edge source="14" target="4" />
+    <edge source="14" target="15" />
+    <edge source="15" target="5" />
+    <edge source="15" target="14" />
+    <edge source="15" target="25" />
+    <edge source="17" target="7" />
+    <edge source="17" target="18" />
+    <edge source="18" target="8" />
+    <edge source="18" target="17" />
+    <edge source="18" target="28" />
+    <edge source="18" target="19" />
+    <edge source="19" target="9" />
+    <edge source="19" target="18" />
+    <edge source="20" target="30" />
+    <edge source="21" target="11" />
+    <edge source="21" target="31" />
+    <edge source="22" target="12" />
+    <edge source="22" target="32" />
+    <edge source="22" target="23" />
+    <edge source="23" target="22" />
+    <edge source="23" target="24" />
+    <edge source="24" target="23" />
+    <edge source="24" target="34" />
+    <edge source="24" target="25" />
+    <edge source="25" target="15" />
+    <edge source="25" target="24" />
+    <edge source="25" target="35" />
+    <edge source="26" target="27" />
+    <edge source="27" target="26" />
+    <edge source="27" target="37" />
+    <edge source="27" target="28" />
+    <edge source="28" target="18" />
+    <edge source="28" target="27" />
+    <edge source="28" target="38" />
+    <edge source="28" target="29" />
+    <edge source="29" target="28" />
+    <edge source="29" target="39" />
+    <edge source="30" target="20" />
+    <edge source="30" target="31" />
+    <edge source="31" target="21" />
+    <edge source="31" target="30" />
+    <edge source="31" target="41" />
+    <edge source="32" target="22" />
+    <edge source="32" target="42" />
+    <edge source="32" target="33" />
+    <edge source="33" target="32" />
+    <edge source="34" target="24" />
+    <edge source="35" target="25" />
+    <edge source="35" target="36" />
+    <edge source="36" target="35" />
+    <edge source="36" target="46" />
+    <edge source="36" target="37" />
+    <edge source="37" target="27" />
+    <edge source="37" target="36" />
+    <edge source="37" target="47" />
+    <edge source="38" target="28" />
+    <edge source="38" target="39" />
+    <edge source="39" target="29" />
+    <edge source="39" target="38" />
+    <edge source="41" target="31" />
+    <edge source="41" target="51" />
+    <edge source="41" target="42" />
+    <edge source="42" target="32" />
+    <edge source="42" target="41" />
+    <edge source="42" target="52" />
+    <edge source="43" target="53" />
+    <edge source="45" target="46" />
+    <edge source="46" target="36" />
+    <edge source="46" target="45" />
+    <edge source="46" target="47" />
+    <edge source="47" target="37" />
+    <edge source="47" target="46" />
+    <edge source="47" target="57" />
+    <edge source="47" target="48" />
+    <edge source="48" target="47" />
+    <edge source="48" target="58" />
+    <edge source="50" target="60" />
+    <edge source="50" target="51" />
+    <edge source="51" target="41" />
+    <edge source="51" target="50" />
+    <edge source="51" target="61" />
+    <edge source="51" target="52" />
+    <edge source="52" target="42" />
+    <edge source="52" target="51" />
+    <edge source="52" target="62" />
+    <edge source="53" target="43" />
+    <edge source="53" target="54" />
+    <edge source="54" target="53" />
+    <edge source="54" target="55" />
+    <edge source="55" target="54" />
+    <edge source="55" target="65" />
+    <edge source="56" target="66" />
+    <edge source="56" target="57" />
+    <edge source="57" target="47" />
+    <edge source="57" target="56" />
+    <edge source="57" target="58" />
+    <edge source="58" target="48" />
+    <edge source="58" target="57" />
+    <edge source="58" target="68" />
+    <edge source="60" target="50" />
+    <edge source="60" target="70" />
+    <edge source="61" target="51" />
+    <edge source="61" target="71" />
+    <edge source="61" target="62" />
+    <edge source="62" target="52" />
+    <edge source="62" target="61" />
+    <edge source="63" target="73" />
+    <edge source="63" target="64" />
+    <edge source="64" target="63" />
+    <edge source="64" target="74" />
+    <edge source="64" target="65" />
+    <edge source="65" target="55" />
+    <edge source="65" target="64" />
+    <edge source="65" target="75" />
+    <edge source="66" target="56" />
+    <edge source="66" target="76" />
+    <edge source="67" target="77" />
+    <edge source="68" target="58" />
+    <edge source="68" target="78" />
+    <edge source="68" target="69" />
+    <edge source="69" target="68" />
+    <edge source="69" target="79" />
+    <edge source="70" target="60" />
+    <edge source="70" target="71" />
+    <edge source="71" target="61" />
+    <edge source="71" target="70" />
+    <edge source="71" target="72" />
+    <edge source="72" target="71" />
+    <edge source="72" target="73" />
+    <edge source="73" target="63" />
+    <edge source="73" target="72" />
+    <edge source="73" target="83" />
+    <edge source="73" target="74" />
+    <edge source="74" target="64" />
+    <edge source="74" target="73" />
+    <edge source="74" target="84" />
+    <edge source="74" target="75" />
+    <edge source="75" target="65" />
+    <edge source="75" target="74" />
+    <edge source="75" target="76" />
+    <edge source="76" target="66" />
+    <edge source="76" target="75" />
+    <edge source="76" target="86" />
+    <edge source="76" target="77" />
+    <edge source="77" target="67" />
+    <edge source="77" target="76" />
+    <edge source="77" target="87" />
+    <edge source="77" target="78" />
+    <edge source="78" target="68" />
+    <edge source="78" target="77" />
+    <edge source="79" target="69" />
+    <edge source="79" target="89" />
+    <edge source="81" target="91" />
+    <edge source="82" target="83" />
+    <edge source="83" target="73" />
+    <edge source="83" target="82" />
+    <edge source="84" target="74" />
+    <edge source="85" target="95" />
+    <edge source="85" target="86" />
+    <edge source="86" target="76" />
+    <edge source="86" target="85" />
+    <edge source="86" target="96" />
+    <edge source="86" target="87" />
+    <edge source="87" target="77" />
+    <edge source="87" target="86" />
+    <edge source="87" target="97" />
+    <edge source="87" target="88" />
+    <edge source="88" target="87" />
+    <edge source="88" target="98" />
+    <edge source="88" target="89" />
+    <edge source="89" target="79" />
+    <edge source="89" target="88" />
+    <edge source="90" target="91" />
+    <edge source="91" target="81" />
+    <edge source="91" target="90" />
+    <edge source="91" target="92" />
+    <edge source="92" target="91" />
+    <edge source="94" target="95" />
+    <edge source="95" target="85" />
+    <edge source="95" target="94" />
+    <edge source="96" target="86" />
+    <edge source="96" target="97" />
+    <edge source="97" target="87" />
+    <edge source="97" target="96" />
+    <edge source="98" target="88" />
+  </graph>
+</graphml>

data/maze/100/maze_graph_C1_RWs.graphml

@@ -0,0 +1,524 @@
+<?xml version='1.0' encoding='utf-8'?>
+<graphml xmlns="http://graphml.graphdrawing.org/xmlns" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://graphml.graphdrawing.org/xmlns http://graphml.graphdrawing.org/xmlns/1.0/graphml.xsd">
+  <key id="d0" for="node" attr.name="label" attr.type="string" />
+  <graph edgedefault="directed">
+    <node id="0">
+      <data key="d0">b</data>
+    </node>
+    <node id="1">
+      <data key="d0">j</data>
+    </node>
+    <node id="2">
+      <data key="d0">h</data>
+    </node>
+    <node id="3">
+      <data key="d0">h</data>
+    </node>
+    <node id="4">
+      <data key="d0">h</data>
+    </node>
+    <node id="5">
+      <data key="d0">j</data>
+    </node>
+    <node id="6">
+      <data key="d0">f</data>
+    </node>
+    <node id="7">
+      <data key="d0">e</data>
+    </node>
+    <node id="8">
+      <data key="d0">f</data>
+    </node>
+    <node id="9">
+      <data key="d0">j</data>
+    </node>
+    <node id="10">
+      <data key="d0">f</data>
+    </node>
+    <node id="11">
+      <data key="d0">i</data>
+    </node>
+    <node id="12">
+      <data key="d0">a</data>
+    </node>
+    <node id="13">
+      <data key="d0">b</data>
+    </node>
+    <node id="14">
+      <data key="d0">e</data>
+    </node>
+    <node id="15">
+      <data key="d0">a</data>
+    </node>
+    <node id="16">
+      <data key="d0">a</data>
+    </node>
+    <node id="17">
+      <data key="d0">c</data>
+    </node>
+    <node id="18">
+      <data key="d0">c</data>
+    </node>
+    <node id="19">
+      <data key="d0">f</data>
+    </node>
+    <node id="20">
+      <data key="d0">i</data>
+    </node>
+    <node id="21">
+      <data key="d0">d</data>
+    </node>
+    <node id="22">
+      <data key="d0">c</data>
+    </node>
+    <node id="23">
+      <data key="d0">b</data>
+    </node>
+    <node id="24">
+      <data key="d0">h</data>
+    </node>
+    <node id="25">
+      <data key="d0">g</data>
+    </node>
+    <node id="26">
+      <data key="d0">b</data>
+    </node>
+    <node id="27">
+      <data key="d0">b</data>
+    </node>
+    <node id="28">
+      <data key="d0">i</data>
+    </node>
+    <node id="29">
+      <data key="d0">a</data>
+    </node>
+    <node id="30">
+      <data key="d0">c</data>
+    </node>
+    <node id="31">
+      <data key="d0">f</data>
+    </node>
+    <node id="32">
+      <data key="d0">c</data>
+    </node>
+    <node id="33">
+      <data key="d0">i</data>
+    </node>
+    <node id="34">
+      <data key="d0">f</data>
+    </node>
+    <node id="35">
+      <data key="d0">h</data>
+    </node>
+    <node id="36">
+      <data key="d0">f</data>
+    </node>
+    <node id="37">
+      <data key="d0">b</data>
+    </node>
+    <node id="38">
+      <data key="d0">i</data>
+    </node>
+    <node id="39">
+      <data key="d0">c</data>
+    </node>
+    <node id="40">
+      <data key="d0">a</data>
+    </node>
+    <node id="41">
+      <data key="d0">a</data>
+    </node>
+    <node id="42">
+      <data key="d0">j</data>
+    </node>
+    <node id="43">
+      <data key="d0">e</data>
+    </node>
+    <node id="44">
+      <data key="d0">g</data>
+    </node>
+    <node id="45">
+      <data key="d0">b</data>
+    </node>
+    <node id="46">
+      <data key="d0">a</data>
+    </node>
+    <node id="47">
+      <data key="d0">a</data>
+    </node>
+    <node id="48">
+      <data key="d0">a</data>
+    </node>
+    <node id="49">
+      <data key="d0">i</data>
+    </node>
+    <node id="50">
+      <data key="d0">d</data>
+    </node>
+    <node id="51">
+      <data key="d0">g</data>
+    </node>
+    <node id="52">
+      <data key="d0">d</data>
+    </node>
+    <node id="53">
+      <data key="d0">g</data>
+    </node>
+    <node id="54">
+      <data key="d0">i</data>
+    </node>
+    <node id="55">
+      <data key="d0">e</data>
+    </node>
+    <node id="56">
+      <data key="d0">a</data>
+    </node>
+    <node id="57">
+      <data key="d0">a</data>
+    </node>
+    <node id="58">
+      <data key="d0">e</data>
+    </node>
+    <node id="59">
+      <data key="d0">d</data>
+    </node>
+    <node id="60">
+      <data key="d0">b</data>
+    </node>
+    <node id="61">
+      <data key="d0">i</data>
+    </node>
+    <node id="62">
+      <data key="d0">f</data>
+    </node>
+    <node id="63">
+      <data key="d0">a</data>
+    </node>
+    <node id="64">
+      <data key="d0">c</data>
+    </node>
+    <node id="65">
+      <data key="d0">f</data>
+    </node>
+    <node id="66">
+      <data key="d0">i</data>
+    </node>
+    <node id="67">
+      <data key="d0">d</data>
+    </node>
+    <node id="68">
+      <data key="d0">d</data>
+    </node>
+    <node id="69">
+      <data key="d0">j</data>
+    </node>
+    <node id="70">
+      <data key="d0">b</data>
+    </node>
+    <node id="71">
+      <data key="d0">e</data>
+    </node>
+    <node id="72">
+      <data key="d0">j</data>
+    </node>
+    <node id="73">
+      <data key="d0">a</data>
+    </node>
+    <node id="74">
+      <data key="d0">d</data>
+    </node>
+    <node id="75">
+      <data key="d0">j</data>
+    </node>
+    <node id="76">
+      <data key="d0">g</data>
+    </node>
+    <node id="77">
+      <data key="d0">g</data>
+    </node>
+    <node id="78">
+      <data key="d0">a</data>
+    </node>
+    <node id="79">
+      <data key="d0">d</data>
+    </node>
+    <node id="80">
+      <data key="d0">d</data>
+    </node>
+    <node id="81">
+      <data key="d0">i</data>
+    </node>
+    <node id="82">
+      <data key="d0">f</data>
+    </node>
+    <node id="83">
+      <data key="d0">d</data>
+    </node>
+    <node id="84">
+      <data key="d0">h</data>
+    </node>
+    <node id="85">
+      <data key="d0">g</data>
+    </node>
+    <node id="86">
+      <data key="d0">i</data>
+    </node>
+    <node id="87">
+      <data key="d0">c</data>
+    </node>
+    <node id="88">
+      <data key="d0">b</data>
+    </node>
+    <node id="89">
+      <data key="d0">f</data>
+    </node>
+    <node id="90">
+      <data key="d0">f</data>
+    </node>
+    <node id="91">
+      <data key="d0">d</data>
+    </node>
+    <node id="92">
+      <data key="d0">h</data>
+    </node>
+    <node id="93">
+      <data key="d0">e</data>
+    </node>
+    <node id="94">
+      <data key="d0">g</data>
+    </node>
+    <node id="95">
+      <data key="d0">f</data>
+    </node>
+    <node id="96">
+      <data key="d0">j</data>
+    </node>
+    <node id="97">
+      <data key="d0">h</data>
+    </node>
+    <node id="98">
+      <data key="d0">e</data>
+    </node>
+    <node id="99">
+      <data key="d0">h</data>
+    </node>
+    <edge source="1" target="11" />
+    <edge source="1" target="2" />
+    <edge source="2" target="1" />
+    <edge source="3" target="4" />
+    <edge source="4" target="3" />
+    <edge source="4" target="5" />
+    <edge source="5" target="4" />
+    <edge source="5" target="6" />
+    <edge source="6" target="5" />
+    <edge source="6" target="7" />
+    <edge source="7" target="6" />
+    <edge source="7" target="8" />
+    <edge source="8" target="7" />
+    <edge source="8" target="18" />
+    <edge source="8" target="9" />
+    <edge source="9" target="8" />
+    <edge source="10" target="11" />
+    <edge source="11" target="1" />
+    <edge source="11" target="10" />
+    <edge source="11" target="12" />
+    <edge source="12" target="11" />
+    <edge source="12" target="22" />
+    <edge source="12" target="13" />
+    <edge source="13" target="12" />
+    <edge source="13" target="23" />
+    <edge source="13" target="14" />
+    <edge source="14" target="13" />
+    <edge source="14" target="24" />
+    <edge source="14" target="15" />
+    <edge source="15" target="14" />
+    <edge source="15" target="16" />
+    <edge source="16" target="15" />
+    <edge source="16" target="26" />
+    <edge source="18" target="8" />
+    <edge source="18" target="28" />
+    <edge source="18" target="19" />
+    <edge source="19" target="18" />
+    <edge source="19" target="29" />
+    <edge source="20" target="30" />
+    <edge source="21" target="31" />
+    <edge source="21" target="22" />
+    <edge source="22" target="12" />
+    <edge source="22" target="21" />
+    <edge source="22" target="32" />
+    <edge source="22" target="23" />
+    <edge source="23" target="13" />
+    <edge source="23" target="22" />
+    <edge source="23" target="33" />
+    <edge source="23" target="24" />
+    <edge source="24" target="14" />
+    <edge source="24" target="23" />
+    <edge source="24" target="25" />
+    <edge source="25" target="24" />
+    <edge source="25" target="35" />
+    <edge source="25" target="26" />
+    <edge source="26" target="16" />
+    <edge source="26" target="25" />
+    <edge source="26" target="27" />
+    <edge source="27" target="26" />
+    <edge source="27" target="37" />
+    <edge source="27" target="28" />
+    <edge source="28" target="18" />
+    <edge source="28" target="27" />
+    <edge source="28" target="29" />
+    <edge source="29" target="19" />
+    <edge source="29" target="28" />
+    <edge source="29" target="39" />
+    <edge source="30" target="20" />
+    <edge source="30" target="40" />
+    <edge source="30" target="31" />
+    <edge source="31" target="21" />
+    <edge source="31" target="30" />
+    <edge source="31" target="32" />
+    <edge source="32" target="22" />
+    <edge source="32" target="31" />
+    <edge source="32" target="33" />
+    <edge source="33" target="23" />
+    <edge source="33" target="32" />
+    <edge source="35" target="25" />
+    <edge source="35" target="45" />
+    <edge source="35" target="36" />
+    <edge source="36" target="35" />
+    <edge source="36" target="37" />
+    <edge source="37" target="27" />
+    <edge source="37" target="36" />
+    <edge source="37" target="47" />
+    <edge source="38" target="39" />
+    <edge source="39" target="29" />
+    <edge source="39" target="38" />
+    <edge source="40" target="30" />
+    <edge source="41" target="42" />
+    <edge source="42" target="41" />
+    <edge source="42" target="43" />
+    <edge source="43" target="42" />
+    <edge source="43" target="44" />
+    <edge source="44" target="43" />
+    <edge source="44" target="54" />
+    <edge source="44" target="45" />
+    <edge source="45" target="35" />
+    <edge source="45" target="44" />
+    <edge source="45" target="55" />
+    <edge source="46" target="56" />
+    <edge source="47" target="37" />
+    <edge source="47" target="57" />
+    <edge source="48" target="58" />
+    <edge source="48" target="49" />
+    <edge source="49" target="48" />
+    <edge source="49" target="59" />
+    <edge source="50" target="60" />
+    <edge source="50" target="51" />
+    <edge source="51" target="50" />
+    <edge source="51" target="52" />
+    <edge source="52" target="51" />
+    <edge source="52" target="62" />
+    <edge source="52" target="53" />
+    <edge source="53" target="52" />
+    <edge source="53" target="63" />
+    <edge source="54" target="44" />
+    <edge source="54" target="64" />
+    <edge source="54" target="55" />
+    <edge source="55" target="45" />
+    <edge source="55" target="54" />
+    <edge source="55" target="56" />
+    <edge source="56" target="46" />
+    <edge source="56" target="55" />
+    <edge source="56" target="66" />
+    <edge source="57" target="47" />
+    <edge source="57" target="58" />
+    <edge source="58" target="48" />
+    <edge source="58" target="57" />
+    <edge source="58" target="68" />
+    <edge source="58" target="59" />
+    <edge source="59" target="49" />
+    <edge source="59" target="58" />
+    <edge source="60" target="50" />
+    <edge source="61" target="62" />
+    <edge source="62" target="52" />
+    <edge source="62" target="61" />
+    <edge source="62" target="63" />
+    <edge source="63" target="53" />
+    <edge source="63" target="62" />
+    <edge source="63" target="64" />
+    <edge source="64" target="54" />
+    <edge source="64" target="63" />
+    <edge source="64" target="65" />
+    <edge source="65" target="64" />
+    <edge source="65" target="75" />
+    <edge source="65" target="66" />
+    <edge source="66" target="56" />
+    <edge source="66" target="65" />
+    <edge source="68" target="58" />
+    <edge source="68" target="78" />
+    <edge source="68" target="69" />
+    <edge source="69" target="68" />
+    <edge source="70" target="80" />
+    <edge source="70" target="71" />
+    <edge source="71" target="70" />
+    <edge source="71" target="81" />
+    <edge source="72" target="82" />
+    <edge source="72" target="73" />
+    <edge source="73" target="72" />
+    <edge source="73" target="74" />
+    <edge source="74" target="73" />
+    <edge source="74" target="84" />
+    <edge source="74" target="75" />
+    <edge source="75" target="65" />
+    <edge source="75" target="74" />
+    <edge source="75" target="76" />
+    <edge source="76" target="75" />
+    <edge source="76" target="86" />
+    <edge source="76" target="77" />
+    <edge source="77" target="76" />
+    <edge source="78" target="68" />
+    <edge source="78" target="88" />
+    <edge source="79" target="89" />
+    <edge source="80" target="70" />
+    <edge source="80" target="90" />
+    <edge source="80" target="81" />
+    <edge source="81" target="71" />
+    <edge source="81" target="80" />
+    <edge source="81" target="91" />
+    <edge source="81" target="82" />
+    <edge source="82" target="72" />
+    <edge source="82" target="81" />
+    <edge source="82" target="83" />
+    <edge source="83" target="82" />
+    <edge source="83" target="84" />
+    <edge source="84" target="74" />
+    <edge source="84" target="83" />
+    <edge source="84" target="85" />
+    <edge source="85" target="84" />
+    <edge source="85" target="95" />
+    <edge source="85" target="86" />
+    <edge source="86" target="76" />
+    <edge source="86" target="85" />
+    <edge source="86" target="96" />
+    <edge source="88" target="78" />
+    <edge source="88" target="98" />
+    <edge source="88" target="89" />
+    <edge source="89" target="79" />
+    <edge source="89" target="88" />
+    <edge source="89" target="99" />
+    <edge source="90" target="80" />
+    <edge source="91" target="81" />
+    <edge source="93" target="94" />
+    <edge source="94" target="93" />
+    <edge source="94" target="95" />
+    <edge source="95" target="85" />
+    <edge source="95" target="94" />
+    <edge source="95" target="96" />
+    <edge source="96" target="86" />
+    <edge source="96" target="95" />
+    <edge source="96" target="97" />
+    <edge source="97" target="96" />
+    <edge source="98" target="88" />
+    <edge source="98" target="99" />
+    <edge source="99" target="89" />
+    <edge source="99" target="98" />
+  </graph>
+</graphml>

data/maze/100/maze_graph_E1_RWs.graphml

@@ -0,0 +1,532 @@
+<?xml version='1.0' encoding='utf-8'?>
+<graphml xmlns="http://graphml.graphdrawing.org/xmlns" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://graphml.graphdrawing.org/xmlns http://graphml.graphdrawing.org/xmlns/1.0/graphml.xsd">
+  <key id="d0" for="node" attr.name="label" attr.type="string" />
+  <graph edgedefault="directed">
+    <node id="0">
+      <data key="d0">i</data>
+    </node>
+    <node id="1">
+      <data key="d0">e</data>
+    </node>
+    <node id="2">
+      <data key="d0">c</data>
+    </node>
+    <node id="3">
+      <data key="d0">f</data>
+    </node>
+    <node id="4">
+      <data key="d0">a</data>
+    </node>
+    <node id="5">
+      <data key="d0">c</data>
+    </node>
+    <node id="6">
+      <data key="d0">i</data>
+    </node>
+    <node id="7">
+      <data key="d0">j</data>
+    </node>
+    <node id="8">
+      <data key="d0">d</data>
+    </node>
+    <node id="9">
+      <data key="d0">e</data>
+    </node>
+    <node id="10">
+      <data key="d0">c</data>
+    </node>
+    <node id="11">
+      <data key="d0">g</data>
+    </node>
+    <node id="12">
+      <data key="d0">c</data>
+    </node>
+    <node id="13">
+      <data key="d0">j</data>
+    </node>
+    <node id="14">
+      <data key="d0">d</data>
+    </node>
+    <node id="15">
+      <data key="d0">j</data>
+    </node>
+    <node id="16">
+      <data key="d0">i</data>
+    </node>
+    <node id="17">
+      <data key="d0">i</data>
+    </node>
+    <node id="18">
+      <data key="d0">f</data>
+    </node>
+    <node id="19">
+      <data key="d0">c</data>
+    </node>
+    <node id="20">
+      <data key="d0">j</data>
+    </node>
+    <node id="21">
+      <data key="d0">i</data>
+    </node>
+    <node id="22">
+      <data key="d0">b</data>
+    </node>
+    <node id="23">
+      <data key="d0">d</data>
+    </node>
+    <node id="24">
+      <data key="d0">j</data>
+    </node>
+    <node id="25">
+      <data key="d0">e</data>
+    </node>
+    <node id="26">
+      <data key="d0">a</data>
+    </node>
+    <node id="27">
+      <data key="d0">c</data>
+    </node>
+    <node id="28">
+      <data key="d0">h</data>
+    </node>
+    <node id="29">
+      <data key="d0">g</data>
+    </node>
+    <node id="30">
+      <data key="d0">d</data>
+    </node>
+    <node id="31">
+      <data key="d0">g</data>
+    </node>
+    <node id="32">
+      <data key="d0">f</data>
+    </node>
+    <node id="33">
+      <data key="d0">b</data>
+    </node>
+    <node id="34">
+      <data key="d0">c</data>
+    </node>
+    <node id="35">
+      <data key="d0">f</data>
+    </node>
+    <node id="36">
+      <data key="d0">c</data>
+    </node>
+    <node id="37">
+      <data key="d0">i</data>
+    </node>
+    <node id="38">
+      <data key="d0">b</data>
+    </node>
+    <node id="39">
+      <data key="d0">d</data>
+    </node>
+    <node id="40">
+      <data key="d0">i</data>
+    </node>
+    <node id="41">
+      <data key="d0">h</data>
+    </node>
+    <node id="42">
+      <data key="d0">f</data>
+    </node>
+    <node id="43">
+      <data key="d0">h</data>
+    </node>
+    <node id="44">
+      <data key="d0">h</data>
+    </node>
+    <node id="45">
+      <data key="d0">h</data>
+    </node>
+    <node id="46">
+      <data key="d0">h</data>
+    </node>
+    <node id="47">
+      <data key="d0">h</data>
+    </node>
+    <node id="48">
+      <data key="d0">j</data>
+    </node>
+    <node id="49">
+      <data key="d0">b</data>
+    </node>
+    <node id="50">
+      <data key="d0">e</data>
+    </node>
+    <node id="51">
+      <data key="d0">e</data>
+    </node>
+    <node id="52">
+      <data key="d0">d</data>
+    </node>
+    <node id="53">
+      <data key="d0">e</data>
+    </node>
+    <node id="54">
+      <data key="d0">f</data>
+    </node>
+    <node id="55">
+      <data key="d0">d</data>
+    </node>
+    <node id="56">
+      <data key="d0">j</data>
+    </node>
+    <node id="57">
+      <data key="d0">g</data>
+    </node>
+    <node id="58">
+      <data key="d0">b</data>
+    </node>
+    <node id="59">
+      <data key="d0">e</data>
+    </node>
+    <node id="60">
+      <data key="d0">f</data>
+    </node>
+    <node id="61">
+      <data key="d0">a</data>
+    </node>
+    <node id="62">
+      <data key="d0">e</data>
+    </node>
+    <node id="63">
+      <data key="d0">e</data>
+    </node>
+    <node id="64">
+      <data key="d0">f</data>
+    </node>
+    <node id="65">
+      <data key="d0">g</data>
+    </node>
+    <node id="66">
+      <data key="d0">g</data>
+    </node>
+    <node id="67">
+      <data key="d0">h</data>
+    </node>
+    <node id="68">
+      <data key="d0">a</data>
+    </node>
+    <node id="69">
+      <data key="d0">j</data>
+    </node>
+    <node id="70">
+      <data key="d0">h</data>
+    </node>
+    <node id="71">
+      <data key="d0">d</data>
+    </node>
+    <node id="72">
+      <data key="d0">j</data>
+    </node>
+    <node id="73">
+      <data key="d0">f</data>
+    </node>
+    <node id="74">
+      <data key="d0">i</data>
+    </node>
+    <node id="75">
+      <data key="d0">f</data>
+    </node>
+    <node id="76">
+      <data key="d0">i</data>
+    </node>
+    <node id="77">
+      <data key="d0">c</data>
+    </node>
+    <node id="78">
+      <data key="d0">j</data>
+    </node>
+    <node id="79">
+      <data key="d0">i</data>
+    </node>
+    <node id="80">
+      <data key="d0">d</data>
+    </node>
+    <node id="81">
+      <data key="d0">b</data>
+    </node>
+    <node id="82">
+      <data key="d0">e</data>
+    </node>
+    <node id="83">
+      <data key="d0">e</data>
+    </node>
+    <node id="84">
+      <data key="d0">e</data>
+    </node>
+    <node id="85">
+      <data key="d0">f</data>
+    </node>
+    <node id="86">
+      <data key="d0">h</data>
+    </node>
+    <node id="87">
+      <data key="d0">g</data>
+    </node>
+    <node id="88">
+      <data key="d0">i</data>
+    </node>
+    <node id="89">
+      <data key="d0">f</data>
+    </node>
+    <node id="90">
+      <data key="d0">d</data>
+    </node>
+    <node id="91">
+      <data key="d0">e</data>
+    </node>
+    <node id="92">
+      <data key="d0">j</data>
+    </node>
+    <node id="93">
+      <data key="d0">j</data>
+    </node>
+    <node id="94">
+      <data key="d0">c</data>
+    </node>
+    <node id="95">
+      <data key="d0">g</data>
+    </node>
+    <node id="96">
+      <data key="d0">b</data>
+    </node>
+    <node id="97">
+      <data key="d0">f</data>
+    </node>
+    <node id="98">
+      <data key="d0">e</data>
+    </node>
+    <node id="99">
+      <data key="d0">b</data>
+    </node>
+    <edge source="0" target="10" />
+    <edge source="1" target="11" />
+    <edge source="1" target="2" />
+    <edge source="2" target="1" />
+    <edge source="2" target="12" />
+    <edge source="2" target="3" />
+    <edge source="3" target="2" />
+    <edge source="3" target="13" />
+    <edge source="4" target="14" />
+    <edge source="4" target="5" />
+    <edge source="5" target="4" />
+    <edge source="5" target="15" />
+    <edge source="5" target="6" />
+    <edge source="6" target="5" />
+    <edge source="6" target="7" />
+    <edge source="7" target="6" />
+    <edge source="10" target="0" />
+    <edge source="10" target="11" />
+    <edge source="11" target="1" />
+    <edge source="11" target="10" />
+    <edge source="11" target="12" />
+    <edge source="12" target="2" />
+    <edge source="12" target="11" />
+    <edge source="12" target="22" />
+    <edge source="13" target="3" />
+    <edge source="13" target="23" />
+    <edge source="13" target="14" />
+    <edge source="14" target="4" />
+    <edge source="14" target="13" />
+    <edge source="14" target="15" />
+    <edge source="15" target="5" />
+    <edge source="15" target="14" />
+    <edge source="15" target="16" />
+    <edge source="16" target="15" />
+    <edge source="16" target="17" />
+    <edge source="17" target="16" />
+    <edge source="17" target="27" />
+    <edge source="20" target="21" />
+    <edge source="21" target="20" />
+    <edge source="21" target="31" />
+    <edge source="21" target="22" />
+    <edge source="22" target="12" />
+    <edge source="22" target="21" />
+    <edge source="22" target="32" />
+    <edge source="23" target="13" />
+    <edge source="23" target="33" />
+    <edge source="23" target="24" />
+    <edge source="24" target="23" />
+    <edge source="24" target="34" />
+    <edge source="24" target="25" />
+    <edge source="25" target="24" />
+    <edge source="25" target="26" />
+    <edge source="26" target="25" />
+    <edge source="26" target="36" />
+    <edge source="26" target="27" />
+    <edge source="27" target="17" />
+    <edge source="27" target="26" />
+    <edge source="27" target="28" />
+    <edge source="28" target="27" />
+    <edge source="28" target="38" />
+    <edge source="28" target="29" />
+    <edge source="29" target="28" />
+    <edge source="29" target="39" />
+    <edge source="31" target="21" />
+    <edge source="31" target="41" />
+    <edge source="31" target="32" />
+    <edge source="32" target="22" />
+    <edge source="32" target="31" />
+    <edge source="32" target="33" />
+    <edge source="33" target="23" />
+    <edge source="33" target="32" />
+    <edge source="34" target="24" />
+    <edge source="35" target="45" />
+    <edge source="35" target="36" />
+    <edge source="36" target="26" />
+    <edge source="36" target="35" />
+    <edge source="36" target="46" />
+    <edge source="36" target="37" />
+    <edge source="37" target="36" />
+    <edge source="37" target="47" />
+    <edge source="37" target="38" />
+    <edge source="38" target="28" />
+    <edge source="38" target="37" />
+    <edge source="38" target="48" />
+    <edge source="39" target="29" />
+    <edge source="40" target="50" />
+    <edge source="40" target="41" />
+    <edge source="41" target="31" />
+    <edge source="41" target="40" />
+    <edge source="41" target="51" />
+    <edge source="41" target="42" />
+    <edge source="42" target="41" />
+    <edge source="42" target="43" />
+    <edge source="43" target="42" />
+    <edge source="43" target="53" />
+    <edge source="43" target="44" />
+    <edge source="44" target="43" />
+    <edge source="44" target="54" />
+    <edge source="45" target="35" />
+    <edge source="45" target="55" />
+    <edge source="46" target="36" />
+    <edge source="46" target="56" />
+    <edge source="47" target="37" />
+    <edge source="47" target="57" />
+    <edge source="47" target="48" />
+    <edge source="48" target="38" />
+    <edge source="48" target="47" />
+    <edge source="48" target="58" />
+    <edge source="48" target="49" />
+    <edge source="49" target="48" />
+    <edge source="49" target="59" />
+    <edge source="50" target="40" />
+    <edge source="50" target="51" />
+    <edge source="51" target="41" />
+    <edge source="51" target="50" />
+    <edge source="51" target="61" />
+    <edge source="51" target="52" />
+    <edge source="52" target="51" />
+    <edge source="52" target="62" />
+    <edge source="53" target="43" />
+    <edge source="53" target="63" />
+    <edge source="54" target="44" />
+    <edge source="54" target="64" />
+    <edge source="54" target="55" />
+    <edge source="55" target="45" />
+    <edge source="55" target="54" />
+    <edge source="55" target="65" />
+    <edge source="55" target="56" />
+    <edge source="56" target="46" />
+    <edge source="56" target="55" />
+    <edge source="56" target="57" />
+    <edge source="57" target="47" />
+    <edge source="57" target="56" />
+    <edge source="57" target="67" />
+    <edge source="57" target="58" />
+    <edge source="58" target="48" />
+    <edge source="58" target="57" />
+    <edge source="58" target="68" />
+    <edge source="58" target="59" />
+    <edge source="59" target="49" />
+    <edge source="59" target="58" />
+    <edge source="59" target="69" />
+    <edge source="60" target="70" />
+    <edge source="61" target="51" />
+    <edge source="61" target="71" />
+    <edge source="61" target="62" />
+    <edge source="62" target="52" />
+    <edge source="62" target="61" />
+    <edge source="63" target="53" />
+    <edge source="63" target="73" />
+    <edge source="64" target="54" />
+    <edge source="64" target="74" />
+    <edge source="65" target="55" />
+    <edge source="66" target="67" />
+    <edge source="67" target="57" />
+    <edge source="67" target="66" />
+    <edge source="67" target="77" />
+    <edge source="67" target="68" />
+    <edge source="68" target="58" />
+    <edge source="68" target="67" />
+    <edge source="69" target="59" />
+    <edge source="69" target="79" />
+    <edge source="70" target="60" />
+    <edge source="70" target="80" />
+    <edge source="70" target="71" />
+    <edge source="71" target="61" />
+    <edge source="71" target="70" />
+    <edge source="71" target="72" />
+    <edge source="72" target="71" />
+    <edge source="73" target="63" />
+    <edge source="73" target="74" />
+    <edge source="74" target="64" />
+    <edge source="74" target="73" />
+    <edge source="74" target="84" />
+    <edge source="76" target="77" />
+    <edge source="77" target="67" />
+    <edge source="77" target="76" />
+    <edge source="77" target="78" />
+    <edge source="78" target="77" />
+    <edge source="78" target="88" />
+    <edge source="79" target="69" />
+    <edge source="80" target="70" />
+    <edge source="80" target="90" />
+    <edge source="80" target="81" />
+    <edge source="81" target="80" />
+    <edge source="81" target="91" />
+    <edge source="81" target="82" />
+    <edge source="82" target="81" />
+    <edge source="82" target="92" />
+    <edge source="82" target="83" />
+    <edge source="83" target="82" />
+    <edge source="83" target="93" />
+    <edge source="84" target="74" />
+    <edge source="84" target="94" />
+    <edge source="84" target="85" />
+    <edge source="85" target="84" />
+    <edge source="85" target="95" />
+    <edge source="86" target="96" />
+    <edge source="87" target="97" />
+    <edge source="88" target="78" />
+    <edge source="88" target="98" />
+    <edge source="88" target="89" />
+    <edge source="89" target="88" />
+    <edge source="89" target="99" />
+    <edge source="90" target="80" />
+    <edge source="91" target="81" />
+    <edge source="91" target="92" />
+    <edge source="92" target="82" />
+    <edge source="92" target="91" />
+    <edge source="93" target="83" />
+    <edge source="94" target="84" />
+    <edge source="94" target="95" />
+    <edge source="95" target="85" />
+    <edge source="95" target="94" />
+    <edge source="95" target="96" />
+    <edge source="96" target="86" />
+    <edge source="96" target="95" />
+    <edge source="96" target="97" />
+    <edge source="97" target="87" />
+    <edge source="97" target="96" />
+    <edge source="97" target="98" />
+    <edge source="98" target="88" />
+    <edge source="98" target="97" />
+    <edge source="98" target="99" />
+    <edge source="99" target="89" />
+    <edge source="99" target="98" />
+  </graph>
+</graphml>

data/maze/100/maze_graph_H1_RWs.graphml

@@ -0,0 +1,522 @@
+<?xml version='1.0' encoding='utf-8'?>
+<graphml xmlns="http://graphml.graphdrawing.org/xmlns" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://graphml.graphdrawing.org/xmlns http://graphml.graphdrawing.org/xmlns/1.0/graphml.xsd">
+  <key id="d0" for="node" attr.name="label" attr.type="string" />
+  <graph edgedefault="directed">
+    <node id="0">
+      <data key="d0">e</data>
+    </node>
+    <node id="1">
+      <data key="d0">c</data>
+    </node>
+    <node id="2">
+      <data key="d0">g</data>
+    </node>
+    <node id="3">
+      <data key="d0">c</data>
+    </node>
+    <node id="4">
+      <data key="d0">h</data>
+    </node>
+    <node id="5">
+      <data key="d0">b</data>
+    </node>
+    <node id="6">
+      <data key="d0">h</data>
+    </node>
+    <node id="7">
+      <data key="d0">c</data>
+    </node>
+    <node id="8">
+      <data key="d0">g</data>
+    </node>
+    <node id="9">
+      <data key="d0">f</data>
+    </node>
+    <node id="10">
+      <data key="d0">f</data>
+    </node>
+    <node id="11">
+      <data key="d0">d</data>
+    </node>
+    <node id="12">
+      <data key="d0">f</data>
+    </node>
+    <node id="13">
+      <data key="d0">i</data>
+    </node>
+    <node id="14">
+      <data key="d0">c</data>
+    </node>
+    <node id="15">
+      <data key="d0">a</data>
+    </node>
+    <node id="16">
+      <data key="d0">b</data>
+    </node>
+    <node id="17">
+      <data key="d0">c</data>
+    </node>
+    <node id="18">
+      <data key="d0">d</data>
+    </node>
+    <node id="19">
+      <data key="d0">e</data>
+    </node>
+    <node id="20">
+      <data key="d0">i</data>
+    </node>
+    <node id="21">
+      <data key="d0">f</data>
+    </node>
+    <node id="22">
+      <data key="d0">b</data>
+    </node>
+    <node id="23">
+      <data key="d0">j</data>
+    </node>
+    <node id="24">
+      <data key="d0">f</data>
+    </node>
+    <node id="25">
+      <data key="d0">j</data>
+    </node>
+    <node id="26">
+      <data key="d0">c</data>
+    </node>
+    <node id="27">
+      <data key="d0">h</data>
+    </node>
+    <node id="28">
+      <data key="d0">g</data>
+    </node>
+    <node id="29">
+      <data key="d0">b</data>
+    </node>
+    <node id="30">
+      <data key="d0">e</data>
+    </node>
+    <node id="31">
+      <data key="d0">b</data>
+    </node>
+    <node id="32">
+      <data key="d0">g</data>
+    </node>
+    <node id="33">
+      <data key="d0">i</data>
+    </node>
+    <node id="34">
+      <data key="d0">b</data>
+    </node>
+    <node id="35">
+      <data key="d0">f</data>
+    </node>
+    <node id="36">
+      <data key="d0">j</data>
+    </node>
+    <node id="37">
+      <data key="d0">g</data>
+    </node>
+    <node id="38">
+      <data key="d0">b</data>
+    </node>
+    <node id="39">
+      <data key="d0">j</data>
+    </node>
+    <node id="40">
+      <data key="d0">a</data>
+    </node>
+    <node id="41">
+      <data key="d0">c</data>
+    </node>
+    <node id="42">
+      <data key="d0">g</data>
+    </node>
+    <node id="43">
+      <data key="d0">e</data>
+    </node>
+    <node id="44">
+      <data key="d0">e</data>
+    </node>
+    <node id="45">
+      <data key="d0">d</data>
+    </node>
+    <node id="46">
+      <data key="d0">j</data>
+    </node>
+    <node id="47">
+      <data key="d0">c</data>
+    </node>
+    <node id="48">
+      <data key="d0">b</data>
+    </node>
+    <node id="49">
+      <data key="d0">c</data>
+    </node>
+    <node id="50">
+      <data key="d0">i</data>
+    </node>
+    <node id="51">
+      <data key="d0">e</data>
+    </node>
+    <node id="52">
+      <data key="d0">h</data>
+    </node>
+    <node id="53">
+      <data key="d0">b</data>
+    </node>
+    <node id="54">
+      <data key="d0">a</data>
+    </node>
+    <node id="55">
+      <data key="d0">a</data>
+    </node>
+    <node id="56">
+      <data key="d0">h</data>
+    </node>
+    <node id="57">
+      <data key="d0">j</data>
+    </node>
+    <node id="58">
+      <data key="d0">a</data>
+    </node>
+    <node id="59">
+      <data key="d0">f</data>
+    </node>
+    <node id="60">
+      <data key="d0">c</data>
+    </node>
+    <node id="61">
+      <data key="d0">a</data>
+    </node>
+    <node id="62">
+      <data key="d0">g</data>
+    </node>
+    <node id="63">
+      <data key="d0">j</data>
+    </node>
+    <node id="64">
+      <data key="d0">c</data>
+    </node>
+    <node id="65">
+      <data key="d0">b</data>
+    </node>
+    <node id="66">
+      <data key="d0">e</data>
+    </node>
+    <node id="67">
+      <data key="d0">e</data>
+    </node>
+    <node id="68">
+      <data key="d0">d</data>
+    </node>
+    <node id="69">
+      <data key="d0">b</data>
+    </node>
+    <node id="70">
+      <data key="d0">b</data>
+    </node>
+    <node id="71">
+      <data key="d0">f</data>
+    </node>
+    <node id="72">
+      <data key="d0">g</data>
+    </node>
+    <node id="73">
+      <data key="d0">d</data>
+    </node>
+    <node id="74">
+      <data key="d0">g</data>
+    </node>
+    <node id="75">
+      <data key="d0">i</data>
+    </node>
+    <node id="76">
+      <data key="d0">j</data>
+    </node>
+    <node id="77">
+      <data key="d0">b</data>
+    </node>
+    <node id="78">
+      <data key="d0">e</data>
+    </node>
+    <node id="79">
+      <data key="d0">d</data>
+    </node>
+    <node id="80">
+      <data key="d0">e</data>
+    </node>
+    <node id="81">
+      <data key="d0">h</data>
+    </node>
+    <node id="82">
+      <data key="d0">e</data>
+    </node>
+    <node id="83">
+      <data key="d0">e</data>
+    </node>
+    <node id="84">
+      <data key="d0">i</data>
+    </node>
+    <node id="85">
+      <data key="d0">j</data>
+    </node>
+    <node id="86">
+      <data key="d0">b</data>
+    </node>
+    <node id="87">
+      <data key="d0">a</data>
+    </node>
+    <node id="88">
+      <data key="d0">i</data>
+    </node>
+    <node id="89">
+      <data key="d0">d</data>
+    </node>
+    <node id="90">
+      <data key="d0">h</data>
+    </node>
+    <node id="91">
+      <data key="d0">c</data>
+    </node>
+    <node id="92">
+      <data key="d0">e</data>
+    </node>
+    <node id="93">
+      <data key="d0">d</data>
+    </node>
+    <node id="94">
+      <data key="d0">b</data>
+    </node>
+    <node id="95">
+      <data key="d0">g</data>
+    </node>
+    <node id="96">
+      <data key="d0">j</data>
+    </node>
+    <node id="97">
+      <data key="d0">b</data>
+    </node>
+    <node id="98">
+      <data key="d0">f</data>
+    </node>
+    <node id="99">
+      <data key="d0">j</data>
+    </node>
+    <edge source="1" target="11" />
+    <edge source="1" target="2" />
+    <edge source="2" target="1" />
+    <edge source="2" target="3" />
+    <edge source="3" target="2" />
+    <edge source="3" target="4" />
+    <edge source="4" target="3" />
+    <edge source="4" target="14" />
+    <edge source="7" target="17" />
+    <edge source="7" target="8" />
+    <edge source="8" target="7" />
+    <edge source="9" target="19" />
+    <edge source="10" target="20" />
+    <edge source="10" target="11" />
+    <edge source="11" target="1" />
+    <edge source="11" target="10" />
+    <edge source="11" target="12" />
+    <edge source="12" target="11" />
+    <edge source="12" target="13" />
+    <edge source="13" target="12" />
+    <edge source="13" target="23" />
+    <edge source="13" target="14" />
+    <edge source="14" target="4" />
+    <edge source="14" target="13" />
+    <edge source="14" target="15" />
+    <edge source="15" target="14" />
+    <edge source="15" target="25" />
+    <edge source="16" target="26" />
+    <edge source="16" target="17" />
+    <edge source="17" target="7" />
+    <edge source="17" target="16" />
+    <edge source="17" target="27" />
+    <edge source="18" target="28" />
+    <edge source="19" target="9" />
+    <edge source="20" target="10" />
+    <edge source="20" target="30" />
+    <edge source="21" target="22" />
+    <edge source="22" target="21" />
+    <edge source="22" target="32" />
+    <edge source="22" target="23" />
+    <edge source="23" target="13" />
+    <edge source="23" target="22" />
+    <edge source="23" target="24" />
+    <edge source="24" target="23" />
+    <edge source="24" target="34" />
+    <edge source="24" target="25" />
+    <edge source="25" target="15" />
+    <edge source="25" target="24" />
+    <edge source="26" target="16" />
+    <edge source="26" target="36" />
+    <edge source="26" target="27" />
+    <edge source="27" target="17" />
+    <edge source="27" target="26" />
+    <edge source="27" target="37" />
+    <edge source="27" target="28" />
+    <edge source="28" target="18" />
+    <edge source="28" target="27" />
+    <edge source="28" target="38" />
+    <edge source="28" target="29" />
+    <edge source="29" target="28" />
+    <edge source="30" target="20" />
+    <edge source="30" target="40" />
+    <edge source="30" target="31" />
+    <edge source="31" target="30" />
+    <edge source="32" target="22" />
+    <edge source="32" target="33" />
+    <edge source="33" target="32" />
+    <edge source="33" target="43" />
+    <edge source="33" target="34" />
+    <edge source="34" target="24" />
+    <edge source="34" target="33" />
+    <edge source="34" target="35" />
+    <edge source="35" target="34" />
+    <edge source="35" target="45" />
+    <edge source="35" target="36" />
+    <edge source="36" target="26" />
+    <edge source="36" target="35" />
+    <edge source="36" target="46" />
+    <edge source="37" target="27" />
+    <edge source="37" target="47" />
+    <edge source="38" target="28" />
+    <edge source="38" target="48" />
+    <edge source="38" target="39" />
+    <edge source="39" target="38" />
+    <edge source="39" target="49" />
+    <edge source="40" target="30" />
+    <edge source="40" target="41" />
+    <edge source="41" target="40" />
+    <edge source="42" target="52" />
+    <edge source="42" target="43" />
+    <edge source="43" target="33" />
+    <edge source="43" target="42" />
+    <edge source="43" target="53" />
+    <edge source="43" target="44" />
+    <edge source="44" target="43" />
+    <edge source="44" target="45" />
+    <edge source="45" target="35" />
+    <edge source="45" target="44" />
+    <edge source="46" target="36" />
+    <edge source="46" target="56" />
+    <edge source="46" target="47" />
+    <edge source="47" target="37" />
+    <edge source="47" target="46" />
+    <edge source="47" target="48" />
+    <edge source="48" target="38" />
+    <edge source="48" target="47" />
+    <edge source="48" target="58" />
+    <edge source="48" target="49" />
+    <edge source="49" target="39" />
+    <edge source="49" target="48" />
+    <edge source="49" target="59" />
+    <edge source="50" target="51" />
+    <edge source="51" target="50" />
+    <edge source="51" target="61" />
+    <edge source="51" target="52" />
+    <edge source="52" target="42" />
+    <edge source="52" target="51" />
+    <edge source="53" target="43" />
+    <edge source="53" target="63" />
+    <edge source="53" target="54" />
+    <edge source="54" target="53" />
+    <edge source="55" target="65" />
+    <edge source="56" target="46" />
+    <edge source="56" target="57" />
+    <edge source="57" target="56" />
+    <edge source="57" target="67" />
+    <edge source="57" target="58" />
+    <edge source="58" target="48" />
+    <edge source="58" target="57" />
+    <edge source="58" target="59" />
+    <edge source="59" target="49" />
+    <edge source="59" target="58" />
+    <edge source="60" target="70" />
+    <edge source="61" target="51" />
+    <edge source="63" target="53" />
+    <edge source="63" target="64" />
+    <edge source="64" target="63" />
+    <edge source="65" target="55" />
+    <edge source="65" target="66" />
+    <edge source="66" target="65" />
+    <edge source="67" target="57" />
+    <edge source="67" target="77" />
+    <edge source="68" target="69" />
+    <edge source="69" target="68" />
+    <edge source="69" target="79" />
+    <edge source="70" target="60" />
+    <edge source="70" target="80" />
+    <edge source="70" target="71" />
+    <edge source="71" target="70" />
+    <edge source="72" target="82" />
+    <edge source="72" target="73" />
+    <edge source="73" target="72" />
+    <edge source="73" target="74" />
+    <edge source="74" target="73" />
+    <edge source="74" target="84" />
+    <edge source="74" target="75" />
+    <edge source="75" target="74" />
+    <edge source="75" target="85" />
+    <edge source="77" target="67" />
+    <edge source="77" target="87" />
+    <edge source="77" target="78" />
+    <edge source="78" target="77" />
+    <edge source="79" target="69" />
+    <edge source="79" target="89" />
+    <edge source="80" target="70" />
+    <edge source="80" target="90" />
+    <edge source="80" target="81" />
+    <edge source="81" target="80" />
+    <edge source="81" target="91" />
+    <edge source="81" target="82" />
+    <edge source="82" target="72" />
+    <edge source="82" target="81" />
+    <edge source="82" target="92" />
+    <edge source="82" target="83" />
+    <edge source="83" target="82" />
+    <edge source="83" target="93" />
+    <edge source="83" target="84" />
+    <edge source="84" target="74" />
+    <edge source="84" target="83" />
+    <edge source="84" target="85" />
+    <edge source="85" target="75" />
+    <edge source="85" target="84" />
+    <edge source="85" target="86" />
+    <edge source="86" target="85" />
+    <edge source="86" target="96" />
+    <edge source="86" target="87" />
+    <edge source="87" target="77" />
+    <edge source="87" target="86" />
+    <edge source="87" target="97" />
+    <edge source="87" target="88" />
+    <edge source="88" target="87" />
+    <edge source="88" target="98" />
+    <edge source="88" target="89" />
+    <edge source="89" target="79" />
+    <edge source="89" target="88" />
+    <edge source="89" target="99" />
+    <edge source="90" target="80" />
+    <edge source="90" target="91" />
+    <edge source="91" target="81" />
+    <edge source="91" target="90" />
+    <edge source="91" target="92" />
+    <edge source="92" target="82" />
+    <edge source="92" target="91" />
+    <edge source="93" target="83" />
+    <edge source="93" target="94" />
+    <edge source="94" target="93" />
+    <edge source="94" target="95" />
+    <edge source="95" target="94" />
+    <edge source="95" target="96" />
+    <edge source="96" target="86" />
+    <edge source="96" target="95" />
+    <edge source="96" target="97" />
+    <edge source="97" target="87" />
+    <edge source="97" target="96" />
+    <edge source="98" target="88" />
+    <edge source="99" target="89" />
+  </graph>
+</graphml>

data/maze/100/maze_graph_I1_RWs.graphml

@@ -0,0 +1,520 @@
+<?xml version='1.0' encoding='utf-8'?>
+<graphml xmlns="http://graphml.graphdrawing.org/xmlns" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://graphml.graphdrawing.org/xmlns http://graphml.graphdrawing.org/xmlns/1.0/graphml.xsd">
+  <key id="d0" for="node" attr.name="label" attr.type="string" />
+  <graph edgedefault="directed">
+    <node id="0">
+      <data key="d0">c</data>
+    </node>
+    <node id="1">
+      <data key="d0">i</data>
+    </node>
+    <node id="2">
+      <data key="d0">d</data>
+    </node>
+    <node id="3">
+      <data key="d0">c</data>
+    </node>
+    <node id="4">
+      <data key="d0">h</data>
+    </node>
+    <node id="5">
+      <data key="d0">h</data>
+    </node>
+    <node id="6">
+      <data key="d0">e</data>
+    </node>
+    <node id="7">
+      <data key="d0">a</data>
+    </node>
+    <node id="8">
+      <data key="d0">f</data>
+    </node>
+    <node id="9">
+      <data key="d0">f</data>
+    </node>
+    <node id="10">
+      <data key="d0">c</data>
+    </node>
+    <node id="11">
+      <data key="d0">f</data>
+    </node>
+    <node id="12">
+      <data key="d0">g</data>
+    </node>
+    <node id="13">
+      <data key="d0">f</data>
+    </node>
+    <node id="14">
+      <data key="d0">b</data>
+    </node>
+    <node id="15">
+      <data key="d0">a</data>
+    </node>
+    <node id="16">
+      <data key="d0">j</data>
+    </node>
+    <node id="17">
+      <data key="d0">j</data>
+    </node>
+    <node id="18">
+      <data key="d0">d</data>
+    </node>
+    <node id="19">
+      <data key="d0">f</data>
+    </node>
+    <node id="20">
+      <data key="d0">g</data>
+    </node>
+    <node id="21">
+      <data key="d0">h</data>
+    </node>
+    <node id="22">
+      <data key="d0">j</data>
+    </node>
+    <node id="23">
+      <data key="d0">j</data>
+    </node>
+    <node id="24">
+      <data key="d0">d</data>
+    </node>
+    <node id="25">
+      <data key="d0">j</data>
+    </node>
+    <node id="26">
+      <data key="d0">e</data>
+    </node>
+    <node id="27">
+      <data key="d0">c</data>
+    </node>
+    <node id="28">
+      <data key="d0">b</data>
+    </node>
+    <node id="29">
+      <data key="d0">i</data>
+    </node>
+    <node id="30">
+      <data key="d0">d</data>
+    </node>
+    <node id="31">
+      <data key="d0">d</data>
+    </node>
+    <node id="32">
+      <data key="d0">f</data>
+    </node>
+    <node id="33">
+      <data key="d0">e</data>
+    </node>
+    <node id="34">
+      <data key="d0">i</data>
+    </node>
+    <node id="35">
+      <data key="d0">a</data>
+    </node>
+    <node id="36">
+      <data key="d0">j</data>
+    </node>
+    <node id="37">
+      <data key="d0">c</data>
+    </node>
+    <node id="38">
+      <data key="d0">i</data>
+    </node>
+    <node id="39">
+      <data key="d0">e</data>
+    </node>
+    <node id="40">
+      <data key="d0">b</data>
+    </node>
+    <node id="41">
+      <data key="d0">d</data>
+    </node>
+    <node id="42">
+      <data key="d0">j</data>
+    </node>
+    <node id="43">
+      <data key="d0">d</data>
+    </node>
+    <node id="44">
+      <data key="d0">e</data>
+    </node>
+    <node id="45">
+      <data key="d0">d</data>
+    </node>
+    <node id="46">
+      <data key="d0">e</data>
+    </node>
+    <node id="47">
+      <data key="d0">d</data>
+    </node>
+    <node id="48">
+      <data key="d0">f</data>
+    </node>
+    <node id="49">
+      <data key="d0">d</data>
+    </node>
+    <node id="50">
+      <data key="d0">g</data>
+    </node>
+    <node id="51">
+      <data key="d0">h</data>
+    </node>
+    <node id="52">
+      <data key="d0">j</data>
+    </node>
+    <node id="53">
+      <data key="d0">e</data>
+    </node>
+    <node id="54">
+      <data key="d0">f</data>
+    </node>
+    <node id="55">
+      <data key="d0">d</data>
+    </node>
+    <node id="56">
+      <data key="d0">b</data>
+    </node>
+    <node id="57">
+      <data key="d0">d</data>
+    </node>
+    <node id="58">
+      <data key="d0">h</data>
+    </node>
+    <node id="59">
+      <data key="d0">a</data>
+    </node>
+    <node id="60">
+      <data key="d0">j</data>
+    </node>
+    <node id="61">
+      <data key="d0">g</data>
+    </node>
+    <node id="62">
+      <data key="d0">e</data>
+    </node>
+    <node id="63">
+      <data key="d0">f</data>
+    </node>
+    <node id="64">
+      <data key="d0">c</data>
+    </node>
+    <node id="65">
+      <data key="d0">j</data>
+    </node>
+    <node id="66">
+      <data key="d0">d</data>
+    </node>
+    <node id="67">
+      <data key="d0">b</data>
+    </node>
+    <node id="68">
+      <data key="d0">a</data>
+    </node>
+    <node id="69">
+      <data key="d0">f</data>
+    </node>
+    <node id="70">
+      <data key="d0">h</data>
+    </node>
+    <node id="71">
+      <data key="d0">i</data>
+    </node>
+    <node id="72">
+      <data key="d0">c</data>
+    </node>
+    <node id="73">
+      <data key="d0">d</data>
+    </node>
+    <node id="74">
+      <data key="d0">j</data>
+    </node>
+    <node id="75">
+      <data key="d0">f</data>
+    </node>
+    <node id="76">
+      <data key="d0">e</data>
+    </node>
+    <node id="77">
+      <data key="d0">h</data>
+    </node>
+    <node id="78">
+      <data key="d0">f</data>
+    </node>
+    <node id="79">
+      <data key="d0">j</data>
+    </node>
+    <node id="80">
+      <data key="d0">a</data>
+    </node>
+    <node id="81">
+      <data key="d0">d</data>
+    </node>
+    <node id="82">
+      <data key="d0">e</data>
+    </node>
+    <node id="83">
+      <data key="d0">g</data>
+    </node>
+    <node id="84">
+      <data key="d0">i</data>
+    </node>
+    <node id="85">
+      <data key="d0">g</data>
+    </node>
+    <node id="86">
+      <data key="d0">f</data>
+    </node>
+    <node id="87">
+      <data key="d0">h</data>
+    </node>
+    <node id="88">
+      <data key="d0">d</data>
+    </node>
+    <node id="89">
+      <data key="d0">a</data>
+    </node>
+    <node id="90">
+      <data key="d0">c</data>
+    </node>
+    <node id="91">
+      <data key="d0">e</data>
+    </node>
+    <node id="92">
+      <data key="d0">g</data>
+    </node>
+    <node id="93">
+      <data key="d0">d</data>
+    </node>
+    <node id="94">
+      <data key="d0">h</data>
+    </node>
+    <node id="95">
+      <data key="d0">i</data>
+    </node>
+    <node id="96">
+      <data key="d0">i</data>
+    </node>
+    <node id="97">
+      <data key="d0">d</data>
+    </node>
+    <node id="98">
+      <data key="d0">b</data>
+    </node>
+    <node id="99">
+      <data key="d0">a</data>
+    </node>
+    <edge source="0" target="1" />
+    <edge source="1" target="0" />
+    <edge source="1" target="11" />
+    <edge source="1" target="2" />
+    <edge source="2" target="1" />
+    <edge source="2" target="12" />
+    <edge source="2" target="3" />
+    <edge source="3" target="2" />
+    <edge source="4" target="14" />
+    <edge source="4" target="5" />
+    <edge source="5" target="4" />
+    <edge source="5" target="15" />
+    <edge source="5" target="6" />
+    <edge source="6" target="5" />
+    <edge source="6" target="7" />
+    <edge source="7" target="6" />
+    <edge source="7" target="17" />
+    <edge source="7" target="8" />
+    <edge source="8" target="7" />
+    <edge source="8" target="18" />
+    <edge source="9" target="19" />
+    <edge source="10" target="11" />
+    <edge source="11" target="1" />
+    <edge source="11" target="10" />
+    <edge source="11" target="21" />
+    <edge source="12" target="2" />
+    <edge source="12" target="22" />
+    <edge source="12" target="13" />
+    <edge source="13" target="12" />
+    <edge source="14" target="4" />
+    <edge source="14" target="15" />
+    <edge source="15" target="5" />
+    <edge source="15" target="14" />
+    <edge source="15" target="25" />
+    <edge source="17" target="7" />
+    <edge source="17" target="18" />
+    <edge source="18" target="8" />
+    <edge source="18" target="17" />
+    <edge source="18" target="28" />
+    <edge source="18" target="19" />
+    <edge source="19" target="9" />
+    <edge source="19" target="18" />
+    <edge source="20" target="30" />
+    <edge source="21" target="11" />
+    <edge source="21" target="31" />
+    <edge source="22" target="12" />
+    <edge source="22" target="32" />
+    <edge source="22" target="23" />
+    <edge source="23" target="22" />
+    <edge source="23" target="24" />
+    <edge source="24" target="23" />
+    <edge source="24" target="34" />
+    <edge source="24" target="25" />
+    <edge source="25" target="15" />
+    <edge source="25" target="24" />
+    <edge source="25" target="35" />
+    <edge source="26" target="27" />
+    <edge source="27" target="26" />
+    <edge source="27" target="37" />
+    <edge source="27" target="28" />
+    <edge source="28" target="18" />
+    <edge source="28" target="27" />
+    <edge source="28" target="38" />
+    <edge source="28" target="29" />
+    <edge source="29" target="28" />
+    <edge source="29" target="39" />
+    <edge source="30" target="20" />
+    <edge source="30" target="31" />
+    <edge source="31" target="21" />
+    <edge source="31" target="30" />
+    <edge source="31" target="41" />
+    <edge source="32" target="22" />
+    <edge source="32" target="42" />
+    <edge source="32" target="33" />
+    <edge source="33" target="32" />
+    <edge source="34" target="24" />
+    <edge source="35" target="25" />
+    <edge source="35" target="36" />
+    <edge source="36" target="35" />
+    <edge source="36" target="46" />
+    <edge source="36" target="37" />
+    <edge source="37" target="27" />
+    <edge source="37" target="36" />
+    <edge source="37" target="47" />
+    <edge source="38" target="28" />
+    <edge source="38" target="39" />
+    <edge source="39" target="29" />
+    <edge source="39" target="38" />
+    <edge source="41" target="31" />
+    <edge source="41" target="51" />
+    <edge source="41" target="42" />
+    <edge source="42" target="32" />
+    <edge source="42" target="41" />
+    <edge source="42" target="52" />
+    <edge source="43" target="53" />
+    <edge source="45" target="46" />
+    <edge source="46" target="36" />
+    <edge source="46" target="45" />
+    <edge source="46" target="47" />
+    <edge source="47" target="37" />
+    <edge source="47" target="46" />
+    <edge source="47" target="57" />
+    <edge source="47" target="48" />
+    <edge source="48" target="47" />
+    <edge source="48" target="58" />
+    <edge source="50" target="60" />
+    <edge source="50" target="51" />
+    <edge source="51" target="41" />
+    <edge source="51" target="50" />
+    <edge source="51" target="61" />
+    <edge source="51" target="52" />
+    <edge source="52" target="42" />
+    <edge source="52" target="51" />
+    <edge source="52" target="62" />
+    <edge source="53" target="43" />
+    <edge source="53" target="54" />
+    <edge source="54" target="53" />
+    <edge source="54" target="55" />
+    <edge source="55" target="54" />
+    <edge source="55" target="65" />
+    <edge source="56" target="66" />
+    <edge source="56" target="57" />
+    <edge source="57" target="47" />
+    <edge source="57" target="56" />
+    <edge source="57" target="58" />
+    <edge source="58" target="48" />
+    <edge source="58" target="57" />
+    <edge source="58" target="68" />
+    <edge source="60" target="50" />
+    <edge source="60" target="70" />
+    <edge source="61" target="51" />
+    <edge source="61" target="71" />
+    <edge source="61" target="62" />
+    <edge source="62" target="52" />
+    <edge source="62" target="61" />
+    <edge source="63" target="73" />
+    <edge source="63" target="64" />
+    <edge source="64" target="63" />
+    <edge source="64" target="74" />
+    <edge source="64" target="65" />
+    <edge source="65" target="55" />
+    <edge source="65" target="64" />
+    <edge source="65" target="75" />
+    <edge source="66" target="56" />
+    <edge source="66" target="76" />
+    <edge source="67" target="77" />
+    <edge source="68" target="58" />
+    <edge source="68" target="78" />
+    <edge source="68" target="69" />
+    <edge source="69" target="68" />
+    <edge source="69" target="79" />
+    <edge source="70" target="60" />
+    <edge source="70" target="71" />
+    <edge source="71" target="61" />
+    <edge source="71" target="70" />
+    <edge source="71" target="72" />
+    <edge source="72" target="71" />
+    <edge source="72" target="73" />
+    <edge source="73" target="63" />
+    <edge source="73" target="72" />
+    <edge source="73" target="83" />
+    <edge source="73" target="74" />
+    <edge source="74" target="64" />
+    <edge source="74" target="73" />
+    <edge source="74" target="84" />
+    <edge source="74" target="75" />
+    <edge source="75" target="65" />
+    <edge source="75" target="74" />
+    <edge source="75" target="76" />
+    <edge source="76" target="66" />
+    <edge source="76" target="75" />
+    <edge source="76" target="86" />
+    <edge source="76" target="77" />
+    <edge source="77" target="67" />
+    <edge source="77" target="76" />
+    <edge source="77" target="87" />
+    <edge source="77" target="78" />
+    <edge source="78" target="68" />
+    <edge source="78" target="77" />
+    <edge source="79" target="69" />
+    <edge source="79" target="89" />
+    <edge source="81" target="91" />
+    <edge source="82" target="83" />
+    <edge source="83" target="73" />
+    <edge source="83" target="82" />
+    <edge source="84" target="74" />
+    <edge source="85" target="95" />
+    <edge source="85" target="86" />
+    <edge source="86" target="76" />
+    <edge source="86" target="85" />
+    <edge source="86" target="96" />
+    <edge source="86" target="87" />
+    <edge source="87" target="77" />
+    <edge source="87" target="86" />
+    <edge source="87" target="97" />
+    <edge source="87" target="88" />
+    <edge source="88" target="87" />
+    <edge source="88" target="98" />
+    <edge source="88" target="89" />
+    <edge source="89" target="79" />
+    <edge source="89" target="88" />
+    <edge source="90" target="91" />
+    <edge source="91" target="81" />
+    <edge source="91" target="90" />
+    <edge source="91" target="92" />
+    <edge source="92" target="91" />
+    <edge source="94" target="95" />
+    <edge source="95" target="85" />
+    <edge source="95" target="94" />
+    <edge source="96" target="86" />
+    <edge source="96" target="97" />
+    <edge source="97" target="87" />
+    <edge source="97" target="96" />
+    <edge source="98" target="88" />
+  </graph>
+</graphml>

data/maze/100/meta_A1_RWs.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f17e1eb7e423deb66bb372367a8555b9bab843efe8cb7a8bde627874ea4d4b2
+size 1787

data/maze/100/meta_C1_RWs.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f17e1eb7e423deb66bb372367a8555b9bab843efe8cb7a8bde627874ea4d4b2
+size 1787

data/maze/100/meta_E1_RWs.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b4ef166a533d020bcb4d06c1b3284e4e4a78af3b0124d94d142dfc7209da8f3
+size 1787

data/maze/100/meta_H1_RWs.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f17e1eb7e423deb66bb372367a8555b9bab843efe8cb7a8bde627874ea4d4b2
+size 1787

data/maze/100/meta_I1_RWs.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7615348444bb5f4988d8e680e19f183b84b80b91268e354df3c641737d88918
+size 1797

data/maze/100/test_C1_RWs_100.txt

@@ -0,0 +1,100 @@
+C 11 24 : F T T R L R R R L T T L L L R L L L R T L T L L T F T F L T R T T R F T F T L T T L L F R R R F L L L L T F T T
+C 18 37 : L R T T T T T L T L T T T R L R R R R R F T T R T T T L F T R R R R F R T T T T T L T T L F T T L
+C 31 23 : F F T T L T R R T T R T T F F T R R F T T T F F T T T R L T F T T
+C 52 64 : T F L T T T T T R F T T R L F L T
+C 63 45 : L T T L L L L T L L R L T R T T L T F L F F T F T F T T T F T T T T T F T F T F T F F F L R F L L F L F R T R T T T
+C 83 80 : F T T T F T T T F F T R T R T T T L L F L T T T R F T F L R L T R F T F L T L L T F T F T T T F T L F F T R R T T T L L T F F T R T L L T F T T R T T R T L T T L F T R T R T T T
+C 31 1 : F F T T L L T F L L L R T T T L T T L F R R T T T T T T T T T T T T T L T T T T L T T F T F R
+C 14 22 : R R L R F F T T R T F T R F F T R L T F R T F T F T T T L L T L T
+C 40 13 : L F T F T R L T L F T F F L T T T R L T T T T R T T L T T T F R R R R R R R F F L L T T L F R R T L L T R R F L T F
+C 28 39 : T T F T F T L T T T L R R T
+C 97 91 : T R T R R R L L F T L T R F F T T L L T F R T T T R L T F R T L T T T R T L L F T R T R L T F T F T T T T T R L T T T L T T T L T F T F
+C 25 20 : T F F L T T L T T T T L L F L L L R F L T T T R L R T L L T T T R R F F T L T R T T L T F
+C 60 61 : L R F F T T T L T L T R L T F
+C 85 82 : T F F F T
+C 39 21 : T T T T T T T T T T L F L L R F R L F L R R R T L L L T R R R R R F T L T R R T T T R T T T R L L L T R R T T T T R F T L R R R F R R R L R R
+C 81 96 : L T F T L L T R R T T T R L T L F F R T R R T L T R R R T T T R T F F T T T T T R F T F
+C 50 62 : F F R L L L F T F T L
+C 98 89 : F L F T F R R T T R L T F R R T L L F T T T F R R F F R T R T T L T R T R L L T T T R R R R T L R T F T L F F L L T F T L T L T R T R R T T R T T T L L F T R T T L L L L R T L L T
+C 83 80 : F L T T L F L R F T T R T T R T L T T T T T R R R
+C 41 93 : F T T T T F T T T T T T F T F T T T F R F R F R R T L R T F L T L F R T R T L R T L T L F T T T L F L T R T R F F
+C 73 52 : T L T T L F L R T F F L R L T T T T T F L T T T T T T L L R T R L R T T F R T L T F L T T L T F L T T T T T R F T T F R T R F T L R T T T L R T L
+C 47 88 : R L F T T T F T R L T L F F T R T
+C 54 75 : R R T F F T T T L R L R T L R
+C 85 76 : R R T F L F R T
+C 37 5 : R T T T T T F R F T R R T T T R L F F T F F F T F F F F T F T F F T F F F F T F F F F T T T F
+C 4 7 : T T F T T F F F F T T T T T T T F F F T T T T T F T T F T T T T R T R T F F F T F F T F T
+C 18 37 : F R F T F L R R T T T T T T T T T L F L L L L L L L T T T R T L L R F T R F T L T T T
+C 25 11 : F T F F F L L L L F T F R R R F T R L L F L F T L F R F T F L F F T F L T R L T F L F R T
+C 76 84 : T F T F T F T F F T L R L T R L R T F F L T R T F F T F R R F F F L T T T T T F R R F T F L T R T T T L T T L T T T R R L F T F T L R T L T T T R F R R T L R F R T T R T T L R F T F L L T T
+C 2 45 : T L T R T T T T T L R T L R T T T L R T T T F T R T T T L R L R T T R T L R F R T L R R F R L F F R R T L F L L R L R T L R T F F L L F L T R F L
+C 96 50 : F T T T F T T R R L R T L R F R R L T R L T R L R L T R F F F T L L F F T T L T R F T F T T L T T T T T R T L T R F T F T F F T L T R T L T L T T F T T T T L T R T L T R T
+C 58 27 : R T L R T T F F T T
+C 15 7 : T T F R L F F T R R R F T T R T L F L T T L L T R F L T T T F T F F F T T T F F R F L L T F R T L L R T L T R R R F T F L L T R T R L
+C 19 24 : R T T R F L F T F R L T L R T L R F L T F T L L R F R T T T T T L L F F R T T L F F F T R L L R T F L T R T L R F L R T L T R T F F L R F F L L F F T F T T F T T T F L
+C 39 56 : L T R T T T T T T T L T T L T T T R T L T L L L R T F L T F T T R T T T T T L L T T F T F L L L R L T T F T L T T F L F L R T L T R R T T F F R L F F T F T F L T T T T T R F R T R T T R L L R F
+C 52 82 : R T R R R F T T T T T L R R R T L T L T F T F F R R F T R T T R F T T T T T F L R T T L T R F F L
+C 31 63 : L T R L T F T T T L L T R T F T T F T F T T R R T T T L L L T L T T L F R T L F T R F F T T F T F F R L F F R R T L R T T T L R T L L F F F F R R T T T L T R F T R F F L R T T R F F
+C 25 23 : T R T L F T F F
+C 96 54 : L L F R R F R F T T T F T R F F T L T L R T L R L T F T T F T R R L F T T T F R T T L T T R F T F T T L T T L R
+C 74 83 : R L T R T L T F F R T L F T F F T F T F R T T L T T L T R R R T F F
+C 9 7 : T L F L R R T L L R T L L L R T F T F L F T F T T R F T T L L L R L F T F T T T T T T T T T F F F T T F T T T F T T T F T T T F T F T T T T T F F F
+C 13 48 : F T L T L F T F R L T R T L F F F R F F T T L L T F F T R T R T L T T T R F
+C 85 96 : F T T R L T F T
+C 39 8 : T T T T T T T T T T L F T T L R R T F T R T
+C 97 83 : T T T F F T L L T T F T T T F F R R R T R T F F T R L L T T R T T T L R L L F L T L T R R T F T R R T L L T F T F R L T T T R T L T F F T T T F T T L R L T F R T T R T L F T F F T L L T T R T R
+C 73 56 : F R R T L T L R L L F R T F R R L F T F T L R L L L L F R T L T F R L T R T F R R T R T F T R F L F L R R L T L T T T T R R F
+C 2 47 : T L L T R R T L T T T R T T T T T L T R T L L T R R T L L F R R F T F L T F T R L R F R L R R T T T R T T F L L T L T T T R T R R T L R T L R
+C 86 54 : L T T T F T T R R L F T F T T T T T T F L T T R T L R F R L F T F F F F T L T R T T T F T R F L T F F F T F L L T F L L R R L T R L R T L L R R L T L R T L F T T
+C 28 39 : T T L R T L L R T F T T T R T R R T L F L L L L L R T T T L R F R T
+C 21 11 : R L F L L R T R L R R T L L R L R T F F T T T L T
+C 41 62 : F T T F F F R R R R T F F F T F F F R L R R L T L F R T R F T T T T T L T L L T L T R T T L L T R T L T L F T F F R T L L T R R R F T T T T T L L F T F R R T T T T T L T T R R R F T L L T R
+C 21 23 : R L T T F T F T F L T T F T R R R T L R T L T R T R T R R T R F F L L F T T T T T F T F L R T T F T T R R F R T L L R F L L L F R R F T L T R T R L R F T F T T F L
+C 42 63 : T T T T F F T T T F F T F F T T R T L F F T T T F F F T L L F R T T T T R F R R L T L R R F F
+C 84 93 : L T T L F L L F T F F L T R T F L T F T F T T T R R T L L T R F F T T F T T F T T F T T R R T L T T T T L F F T L L T T T R L R T R T T R L T T T R T R F F T T T T T T
+C 93 61 : F T T F T T L L F F F F T T L T R R T F T L L T T T L T T F L T R F R T T R F T T R T T T T R T L R L L F R L F T F R R F T L L F L T T T R T L T T T R T T T L T T T T T R F F T T R T T R F
+C 74 43 : T F T F F L R T F T F T F F T T F R T R T F L L F T R L F L F R T F F F T T T F F T F F T F
+C 8 39 : F T T T T T L F R F R T L F L R T R T L R F T F L R L T R F L R
+C 91 86 : L F T R L T T T T T R R T L T F T F R R T T T L T R R R R R R T T L L T L F L L T T T F F R T T T R T F R T L R T T
+C 29 18 : R T L R F R T L L R T L R R T L T L T T F T F F T L F R T L F L F T T F T T T F F F T T T F T R L R T T T T T T F R T L T T F L R T T T
+C 65 28 : T R R L T L R R T L T T L R T R L T R T R R T L F T F T T L T T F T F F R R R R L R F R T T T F R F L T T L L
+C 47 99 : R T T T T L F L T R L F T T F L R R R R T T T L T R R T F T F R R F T F T R R T T R T L F T T T F
+C 82 71 : T F L T T T F R T T R F T L R R
+C 78 18 : L R T R F T R T R F T R T T T T T R R L L T R T T T T T T R T L T R F R F F T T R L R R T L L T T T R T T T
+C 81 74 : L L T T T R L F T R R T L L T T F L T T
+C 77 44 : T F T F T T T L F L T F F T F T F T F F T R F L R R L L T T F T L L L T L F T T T F F T
+C 42 53 : F F R F T T T F L F T T F T T T F F T F F T F F R T T L T L R F R L F L T T T R T T F F R T T T L F F T F T T R L L L T T F F T F T T T T F R R T T R R R T T T L T
+C 31 13 : T T T R T F T R F L F R R R L T T R R R L R F R L R T R F T F F F L T L R R R T L T T L R T L T L T R L R T T T R T T T F F
+C 75 85 : T L R T T F R T L F F T T F L L F T T L R T T T F R L T T F T F R R T F F F R R T T F F F F T T T F T T R T L T R L T R R F T F L T T T L F T L R T R R R
+C 61 54 : F F L T R R R R L F T F T T F T L L F F T F R R T L L T R R T L L F T F R T R T L L F T T L T L T R T R T T R L T T F T R T R T T R L F T R
+C 15 30 : F T F F T T F T F T L F T F L T T T R F T L R R R R T F T F T F T L T R L R T T T T T L L F L L T T F L L T F L T F T L F T R R F T F F F L T R T L R L T T T T R T L R F R R T T F F L T R T
+C 69 58 : T R L R T L R L T R T F F F L T R F F T T F L T
+C 14 28 : R L F T F F L T F T T R F F T T R T L F F T T T F T T T L L F F F L R R F T L R F R T T T T T F T R L R R R T
+C 65 93 : F L F T T T T T R T R R T L T T T R L L R T L R T L F T L F R T T F F T F L T L F T F F F T T T T
+C 61 26 : F F F T F T F F F T F T R T L R R L T T F T F L T R T T F F L T T F R L T T F F T F T T F T T T L R T F L L L T L R R T L T T R T F R L F L L T F F R
+C 97 84 : T T T T T T T T T F T L F T F L T F F T F F T R L L L F T T T T T R F R T L T R T T T T T L T R T T T T T F T F T T T F T F T F T R R F T F L L F L
+C 29 35 : R R T L T R T L T T L F L R F
+C 79 78 : R R L T F
+C 29 33 : T T L L T R T L L T F R T L T T T T L R T T R R T L T F F F T T F R L L F
+C 19 7 : T T T T R T T F R T T T L L T R T L R T T F L
+C 60 53 : L T T R F T T F T T T L L L T T T T L T T L T R
+C 24 45 : T R T L F T T F T T T L T L R R F R R T T L F T T R T L R F T T T T T F L R T T L T L T F T T R R T T T L
+C 63 76 : T T F T R L F F T T L T R T T F F T L T T R T T T F T F T F L T L F R L F T F F F F L R T F T T T R R T T T T T T T T F F L L F T L R T L R T F R T R T F T R T T T L T T T F T T T T T T T L L R R T L
+C 28 16 : T F F F T F L T T L T T T R T T T L T R L R R T T T T T L L T F L T R T L T R L T L R
+C 58 37 : R F T F L R F L T T T
+C 22 59 : T T L R F F F T F T T T F T T T T T T T T T T T T T F L L R F T F R F F T L R F R R F R F T T F L F T L L T R F T R R T L F L L T R
+C 69 29 : T T T T T R T F F L L T F R R F F F R L T T L L R R T T F T F L R L T L T R T R R L L T T T R R F T T R F F R L R R T T T L R T F L R T T R T L L R T L T R F L T F T L L R T T T T
+C 74 80 : T F L L T F F L T F R T T T T T T T L F T R R T R T F R T L R T T T R T T T L T L T R T T T F T R R R T L L F T R T F T F L L T R R F T L L T T T F T
+C 50 23 : R T T T R F T F L T R F F T L R T F F T F T T T L L L T R R L L T R R T R R R R T T L T R T R T F F T T F T T L L R T R T L F F R F T F T L T T T R F F L T F R F F T F L R F F T T T F
+C 21 37 : R L T F L T R T R T F T R L T R T L R F R R F F T L R T T R L L F R R L F T L L F T T T T T F T F L R L R T L R L F T F T R T L T R T R F T T L F T F R F L T
+C 42 86 : F F R T T F R R L F T T F T T T T L T R T L T T T R F T T F T F T T T T F L T R F T T F R R T F F R L R F T T L T F R L R T L R R T T F T T T F F L L F T R L F T F R R T T
+C 74 43 : R L R R F T F T T F L F R T T T F T T F F T T T F L L T R T T T T T T R L R T T T F T T F R R T T T F T R F T T L R R T F F R T L T T L F R R T L L L T R L T R T L T T
+C 78 49 : L T F L L T F R T L L R F F T F T T F L R R R F F T F L R T F T F T L T R R R R R R R R R T T R R F T T F F L R T T T L F T L L T R F R T L F L T L T F R T T L T R T T T L L T
+C 64 76 : L T T T L T T R R T T T F
+C 23 14 : R R R T R R R L T L L L F F T F T F T F R R L T T R L T F T T T L L T T T T T T T L T T T T F T T L R L T R T T F L R R T T R R L T R R L R T L L L L T
+C 96 54 : T R R T F F F T T F R T T T F T L R T T R T T R R L T R L T R L T F R R R T R T T T T T F T L F T R T T T T R T T F F L L R R T
+C 48 78 : F T L T R R T L L F L T R T F T T F F T T L L T R T L T R T T T T T T R T T T T T T F F R T T T L F F L L F T T T R T T L T F
+C 56 81 : L T F R L R T F F T T T F F L L R T T L F T R L L L T R R R R T T F T F T R R L T F T T L T F F T L L R L T R R F R T R L T R L R T T T L R F T L T F T T T F L T T L L T T F F T F F R T L
+C 5 44 : F F T T T F F T F T F T T T F F F R F L L T T T R R T T F T F L R T T T F T F T L F L R T T F L R
+C 98 68 : L F T L T R T F L L L L L T R F T L L T F T L L T F T L L T F T L R T L R R T L L L L L F T F T T R T T R F F T F F T F F R T R L T L R T T R R L L T
+C 30 28 : F F F L L T L T R L R T F L L L R R R L R T F T R T T L L T T R F F R T L T T F F F T R F T F L L F R T F T F R R T T F F T T L L T F T T T T F R T T L F L R R R

data/maze/100/train_A1_RWs_500K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:388cf2c5f45c60ddb2bb5901d804178cd4af45bf27661a19cb0b6fee59408b3c
+size 129000000

data/maze/100/train_A1_RWs_500K.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8cda5424503efbf95fe8762d86b37cc962e838af65d8b3c124585e479f30663
+size 56839817

data/maze/100/train_C1_RWs_100.txt

@@ -0,0 +1,300 @@
+C 42 4 : F T T T F T F T F T T T T T F F F L F L T R F R F F T F T T R T L F T F T T R L T R L F T T T T T F T F F T T T F F F L F L R T L F L T L F F L L R R T F F F F T T F T T T T T
+C 64 66 : T T F R R F F T F T T F F F T L F R R L T F L R T T T F F T T F F L T T T T T R F F L R T T
+C 82 81 : T T T F L T T T F T T T F T R L T T L L T T T R T L T R T T T L L R T F T
+C 14 22 : T T T T T T F T T F R R T L T R T L L F L R F
+C 23 49 : F T T F T F T F F L T L R R T R F T F F T T R F T F F F T F T F T L T R L F T R T T F T T R T T F L R T T F T T T F L T R T L T F T L L T R L R T L L R R R R R
+C 11 14 : T T T T F T T R T L T T T R L F L R T F T R L F L T R L T L F F R T L R R T T T T R F R F
+C 64 42 : T T F T R R F L T T T T T R F R T T L F T F F T F F T F F R T T L T T L T F T L F R R T L L F L T F R L R T L L F T L F T F T T R R L L T R R L L F L T T T T F
+C 96 86 : T T L T R F T F L
+C 61 62 : F L R T T T L
+C 52 36 : F T T R L F R T L T F L F T R T L T T T F T L R L T T T L R F R F T T L T R F F L R L T R L T R T R L T T L F T T R F T F L
+C 12 21 : T F T T T F T T R F T T T F T R T R L T F T L T L T R L
+C 88 89 : R L T R F T L T T L T R T R T L T
+C 44 54 : T T R T R R R R L T R F T F L F F T F T F T T T T T F T T F R
+C 74 75 : T F L T T L T T F L T L R R T L R R R R L R L T F F T F F T F L L R F L F T F R L T L R R T R R L T T F T T L T T T R F F R L L T L R R L R T T T T T L L F T T L T R F T T R T L T F
+C 49 59 : T T R T T T L L T T L
+C 5 39 : T T F T T F T T F R F R F F T T T F F F L T T T F R T L T
+C 23 58 : R R F R T T R R L L T T F L T R L F F T T L L T T T R L R T L R T L R F F L T T T R F T F L R L T L T T F T T F L L T T F T F T T R T T T L T R T L T T L R F F
+C 12 11 : F R T L L R L L T T L F R R F R T T R L L T T L F L F R F F T F F T F F R R T L L T T F T T L L T T T F T L L T F F R T T T L T T F L T R F R R T L L F L L T R T L T F L R R F T T L
+C 46 56 : R R T R T T T
+C 15 14 : F T F F T
+C 64 54 : T F R T R T T T F L L T T T T T T T R R F T F T F T T T T T F F T R L T R L T R T L T T F T T T F T T F F T L L T F T F F L T T T R T L T R F R L L T L F T F T L T T R T T T
+C 73 55 : F R L T R T T L F T F F L L T R L R T T F L T R T L T F T F T F R R L R T T T F T T R R L F T F T F T T T R T T R T R T T T L T R L R R L F T R L F L T T R
+C 13 56 : F R L T R R F T T T F L L F F R R T R T F L F L F R T R R R R T L F R T T R L L T T L F F F T F T R L T R T T R L F T
+C 60 51 : L R T T T L T T T T T T T T T R
+C 45 56 : T T L R F T F T F R F T T T F L T R T L F T F R T T F T T T F L F R T L T R L R L F T T T F T F F T T R T T T R T F T T R F R R T T T R T L T L F T T L R F F F R R F T T T T L R F T T T R L F R T
+C 32 15 : T R R L R T L F L T R L L T T R L T T T F T F T T T T T R F L T T R L R F T T T T T L L R T R F T R L F F L F R R L L R T T T T L T F T L T L T F T L F F T R L L R T L R L T R T R T R F L R
+C 10 42 : F F F R L L R T L R R T L L L F L R L R L T R L R L T T L F T L T L R F L L F L T F R L F T F F R R R L R F F R R L F F T F T T T
+C 11 49 : L T R T L T L R T R T L L T R T L L R F T F L F T L R T R T R R R R R R F L T T T F T T T T T F T T T F T F T T T T T R F T R R L T L L L T T L L F F R T R F T T R F F T F L T R F L F L L T R R T L
+C 7 8 : T F T F F F T F T R T R T T T T T F F T T F F T F F F F T T T
+C 23 15 : F L T T T T L T F T F F L F R R T T R R R F R T L L T T T R L T T T R L R L T L R T T R F T L R F T T T R T T T L L R
+C 62 50 : T T F F T F F T L L F L T T T T T T T R F R L T F T T T T T F T T L T T T T T L F T F T T T F T R R T L R R L L R T F F T T T F T F T F F T L R T F T R L L L L R T T T L L T T T F T T R T L F F
+C 36 38 : F R T F R F R R
+C 48 35 : F T L F T T T R T L L T T T L T T T F R L T T T T L T R R F R T T F L F L T F R T L T R T T T L F F R R T L F L T F R F R F R L T F L L T R T L T F R L T T T L F R L L L T R R R T T L L F T T R L F T
+C 79 78 : R R T L T R R
+C 65 74 : R R L R T F T R R L T T R T T L L T L T F T L F R L F T T T T T F R F R F F T T T L R R R R L T F R R F T L T R T L R R L F R R T F T F F T
+C 20 30 : R F T R T R T F T F T T T T T T T T T T T T T F T F T F T R T T T R T F T F T
+C 2 22 : T T T L T R T T T T T L T T R T F R R T
+C 20 22 : R T T T T L F T F L T F T L F T T F T F R T L L
+C 6 76 : F T T F R L T T R F R T L T T T R T L L R T T F T F T F R T L F R L F T L T L R L T L L T T T F T F T R R R T R L T F T T T F F F T F F T T F T T T F T F T L F R F R T R F T R F F R L L R R T F
+C 5 4 : T T F T F T F T F F T F T F T F F T T T T F T T F R F L T T L L T R R T R T T R T T T L F T R T T T T T L L T T T R R T R T F T R R R R R T L T T T T L L T F L F F F F T T T T T
+C 44 37 : F R T T R T T R L T R F T F L F T T F T F T F T T T T T T T F F T F F T T T F F F R T L L L L L L F L F T T L F T F T F T F T T T F T R R F R T T F
+C 43 42 : T F T T T F T F T T T T T F F T F F T F T F T F T F T T T T T F T F T F F R T R T T T F F T T T T F T F F F T L T T L T R R L T R T T L L F L T R T T F R L T F T L L L L F F T T T T F T F T F T
+C 62 52 : T T F T F T L L F L T T T R T L T T T T T R F
+C 45 96 : T F T R L F L T T T F R L L F T T T T T L R T R L T R T
+C 5 27 : T F T T T T T T T T T F T T F F T F T T T F T F F F T F T F T T T T T T T T T F T T F F F R F T F L F T F T T R F L T F F F T F R F F L T R F T T F R T T T F T F L R R T T F T L F
+C 50 35 : F T L T R T L T T T R F T F T T T F F R T L T T F T T F T T L T T T T T T T T T T T T T R F F T L T L T R R T T T F T F T T T F F T F T F F T T T R R T R L F T F F L T T F T
+C 81 86 : T T T L T R T T F T F R R T T R F T T T R F T T F F T R T T T L F
+C 61 96 : F F F F T T R L F T L R T T T L R T L F T L F T F T F R R F F T T T T T T T F T T F T T L L R L F T F F T F F T T T F R L T T T R L L L L R T F T F T F T T L F L T F T T T F T T T F T L R T R
+C 72 65 : F T T F F T T F T R T L F T F F T L
+C 39 74 : T T T T L L F T T L T L F L T T F F L R R L L F T F F T F F T T T F F F T
+C 23 39 : F F F L L T T T T T R L R T R L T T F T F R T T L T T F F T F L F T T R R L T T T R T T T F T T F F R R T T T L L F T L R R F T L T L T T T R R R T T R T T F T T
+C 36 55 : F L T F T T T T T T F T F L T T F L R F F T T F T F F R T R R L L T R R T
+C 2 30 : T L L T R T L F F T L R R R T F R R T L R T L R T T T L L F T L F L T R L T T L L F L T T L F L T R R T T L L L R T F R F T F T R T T T T T R
+C 42 21 : T T T T T T F F F L R F L L R T T L T R R L L F L T F F L T T T R T L L T T R R T L T T L T R R T T T T T T R L L T R T L T R F T R T T R F F L R L F L R L R F T T R T T T T
+C 36 51 : T L F L L T R F R T F R T T T L T T F T T T F T T R F F T L L F T T L T R F F R R R L
+C 65 50 : R T L R R T R T T R R L R R F F F T T R R R R R L T T F L T R T T F F T F F T T T T L T R T L T R F T F
+C 39 30 : L F T F T F T R F T F R T F T R R T T T R L T R T F T F L T R F T T R T R F R R T F F T T T F T L R L R F R T T T F T
+C 66 54 : T T L F T F T L F L L T F T T F F T T T T T L T T T L F T F F T F F L
+C 29 32 : T T L T R F T L F T L T T R R F T T T T F R T R F F R T R T T L L T
+C 62 63 : F F T F F T L T R T L L T R L T F T L T R T L T L T R T T R T L L T F T L T T R T L R T F L L L T R R T T
+C 30 1 : R T R F L T T R R R T T R F T L F T T T R L F R
+C 48 59 : F T T T L R R F T F L T T F T T L L L T T L L L L R R F T F L F L T R T L T R R T T R T T T T R T L T R T T R F R F L T R F T T L F T L F T F T F T T T R T L F T F R T R T L T L T F L T L R R
+C 8 3 : T F T T F T T T T F T T T F T F F T T T T
+C 38 6 : F L T R T T T L F T R F T L T R T L F R T F F F F F T T T T T F F F T T F R T L F
+C 27 12 : F T T F T T T F L F T L F T T R R T F T F T F R R F R L T R L F L T L L T R R T L F
+C 5 27 : F T T T F F T T T T T T T F T T T F T T T T T F F T T F F F T T T T T L L T R T L R R F T T T T L T R F R T T R T L L F F T T R T L
+C 44 89 : T F T F F L T T R F T F L T F L F T T T T R T L F R L F R T L T R F L R F F F L T T L
+C 4 5 : T T T T F F T T F T T T T F F T F F T T F F F T T T T T F F F F T T T T F T F T T F F T T F T T T
+C 31 30 : F L T R R T L T T F T R F R R F T L L R F R T F T F T T T T T T T F T F T
+C 40 49 : L T T T T R L T R T F F T T T R F R F R R R T L L L L T R T F T F T T F T T T T F T F T F R L T T T T T F L L F L L T F L F R T T T L L T L R L T R T L R T R T T F L L R
+C 47 68 : R L F T F T R L T R F R T L L L L T T L T R T T R T T R L F T
+C 56 50 : R T F T F T L T T R L F T R F L R L T T F T T T T T F F F T F T L R F T T T F L R T F T F F T R L T T R T R F T L L F T R L L L F T T T R T R T T T F T R T R T T R R R L F T T T F R T T L L L F T T F
+C 35 27 : L R L T T T L F F L T R T R R T L T R T T T L L F
+C 66 74 : T L L T T F T T T F F
+C 33 14 : T T T T T T T F F T L R R R R R L R F T F L T T L F L L L F R T L F R T F T T F T F R F T F L F T T R R T L T R T L T T T R T T T T T L L R F T F R R T T F T L T F L L F T F R R T L T T R F T T T F
+C 99 58 : L F T R T T R T F L T T T R F T L T L L T T L F T R R F R T T T L T F F R R R T T R T L L T T L L
+C 58 68 : R L T R T L T T T T T T T L T F R L L L T R T L L L T T T R F T F T R T L L L T T L F L T R R L L T T L R R T T T L L T L L L T R T T T L T R R F T T R T L F T R T T T F
+C 57 11 : F T R T T T L F T R T F T L R F F L R L T T T T F T T F T R T L F F R R T L T R T L T R T L R T L T T T T R T T T T T L L T F T F T F T
+C 8 6 : R T L F F F T T T F T F T T T T F T F T F T T T T T T T T T F F
+C 97 65 : T T T R T R T L F T T L F L T L F T T L T T R R L T T L L F L R F R T T T T R F F F R T L L T R R T T T L F T R T R T T R R T L F T T T T T T T R T L T F T F T T T T T F T F T R F L R T L T R F F L
+C 44 85 : F L R F R T L F L F T L F F T T T T F T F F R L F T R F T F T T L L T T T F T L F L L R R L L R L T T F T T T F R T L F T F T L F L R R T F T T T T R T L R L T F L R T F
+C 80 94 : F T T F L T T R F T T R L R T L T F R R R R T F T R T T L T F F T F F T F F T T T F L T R
+C 58 56 : F L L T R R F R F L F R T F R L T T L F
+C 52 62 : F T L L L T T T T T L L R L L F F T T T F T T T F T T F T F F F L R T L T R T R L T R F F R L L T F T T F R L F T F T F L T T T T T T T T T T T T T R T L T T T R F F T F T F R L T F F T T T
+C 88 89 : R L T R R R R R R L T T T R T T R T L T T T L T R R R R R T R F T F F L L F T
+C 80 79 : L R R L F T F F L T F T T R R F T R T L T R L R F R L T R R T T L L R R T T T R R R T F F L F T F T F F T F T T F F F R F F T T L R L T L F L L T T
+C 65 66 : R R T T F T F F T R T L T R R L T
+C 26 37 : T L T R T L F T F T T T L F T T
+C 26 46 : T L F F L T R L L F T R L T L L T R R T R T F T F T T T F L
+C 49 59 : T L T R T L T T T R R R R T L T T L T T T T T
+C 11 12 : T T L T L F T L L L R R T L F
+C 20 15 : R T T F T F T F T R L T L F L R T R T T L T T F F T F R T T F R F R T F T T T R T L T T T R T T L T R R T L T T T L T R F L T T F R F T L L T R L F T R L T T F L L F T F F F T
+C 32 38 : L F T T L T T R R T L L R R L L F L R F R L T R R L T L F F T F L T L R T T T R T F F F L L T R T L T T L L T F L R F L T R F L L T T F L L F L L T F F T R T L R T L T T T T T T R F T L R T T F R
+C 32 21 : T R T L F T T L L T F L T R L R T T T L T T F L F F T F T T R R T L T T R R F R F T L L L L T T L T L T R T L L R T T T L R R T L T T T T T F F R T F T F T L L
+C 62 64 : T T L T L F
+C 39 9 : T T T T L L F T L T T F R T T T T T T T F T F T F T F T L F T R R R F T F T T T T T R T T F R
+C 24 35 : F T T R T R F F L T T T T R R R F L R F
+C 93 85 : F T T T T F L R T F R T T R T T T L T L F T F F F F T L F T T L T R F L F T L L R T L F T T T F T T T F T L L F T F F F T R T L F L R L T F T T T L L R T T F T T R T T L
+C 47 51 : R T T T F F L F T T F T R F R F T F T T R L T R T T T L T T L T R T R T L F T T R T T L T R F R F F T L R R R R R T L L F T R L T T T R L F T F F T L T R R L
+C 84 54 : T T L R L R T T L F T R F
+C 64 45 : F F L L T R T F T T T R F T L L L L L
+C 22 33 : L L T R F T L L L L T R T F T T T F T F L T
+C 37 7 : T T L R L T L L L L T T T F R T F
+C 16 25 : T T T F L R T L L L L F R T T T T T T T T T
+C 91 81 : L L L T T T R L L L L R T T T
+C 94 85 : T T F T F T T T F F L L L L T F T L L F T F
+C 18 18 : F T T R T T T L L L L T F R T L F L
+C 27 45 : F L T L L L L R F F L L T L R T
+C 59 49 : T R R T T R T L L L L L L L T T L L T
+C 79 37 : R R R F F R L L L L L T R F R F T T T T
+C 37 32 : T F R L F F L L L L L R T
+C 84 74 : F F R T L L L L L F R L F L L T R R F T T
+C 59 69 : L L L T T L L L L F L
+C 81 90 : L L L L L L L F
+C 53 63 : R T L L L L T R F T L T L
+C 19 8 : T L L L L R R T T T L F T F L T T F T F
+C 72 80 : R T R T L R L T F L L L L T T T R L T
+C 86 80 : T L L L L F F F F F T L T R L T T T T T
+C 15 32 : F R R F F R T T L T R L L L L T R R T L T T F L R T R
+C 29 8 : T T R T L T L L L L L L R L T
+C 40 25 : L R L R R T T T L L L L L T L T R T F F T T F T T
+C 89 99 : L T R L L L L T R R R T L T T T T
+C 33 31 : T T T F T F T R R L L L L T F T T L R L T R T R T L
+C 9 29 : T L F L L L L L
+C 61 46 : F T T L T L L L L L T T F L T L F L F
+C 23 14 : L L L L T L T R L T T L T R T R
+C 49 68 : T T T L L L T R T L L L L L T R L L T
+C 95 66 : F T R R T L L L L T R T L T L L R R
+C 55 54 : L L L L T
+C 35 18 : L L T F F F F L L L L R R T L F L
+C 63 61 : T F T F L L L L L L L R
+C 55 44 : L L L L L F F T F F T F T F R R
+C 47 58 : R L F L L T R R T L L L L L L L L L T R
+C 88 89 : R L L L L L L L T L T
+C 33 24 : T R F R R T T L T R L L L L R R
+C 32 32 : L L L L F L L L
+C 88 28 : L F F F T L T R T T T L L L L T T R R F T T F R
+C 93 84 : F T T T T F F L L L L T R L R L F L L F
+C 84 74 : L L T R R T F R L F T F T L L L L L F L F F F T F
+C 78 48 : L F T T R L L L L L L T T T T T R T L T R T L
+C 33 22 : L L L L T R L L T T T R R T
+C 20 21 : R T T L F L L L L T F L T F T T T L F F T F R
+C 53 66 : R R T F T T L R L L L L L L L L F T T R T T
+C 10 30 : F T T F T F T T T F R L L L L T L T R R L R
+C 89 89 : L T F T L L L L
+C 99 78 : T R T L L L L T F
+C 49 58 : T L L L L L
+C 61 51 : F T T F L L L L L L F T T
+C 61 53 : F L R T L L L L F F T T T F F
+C 77 95 : T L F L T T T R L L L L L L T T L T
+C 54 56 : L T L L L L L F
+C 55 56 : L L L L T T F
+C 93 85 : F T T T T F F L L L L F T R F T R
+C 21 32 : R R L T R F L L L L F T T T
+C 40 20 : L T T T T R F F L L L L L R T F L R F R T T T T
+C 94 96 : T T T T F F L L L L F T
+C 53 64 : R T L L L L T L
+C 30 21 : F F L L L L L L
+C 29 37 : L L L L L T R F T F T F L F F T F
+C 44 44 : R L L L L L F T R L F T R L L F T F R R
+C 61 66 : F L T L L L L L F F F L T
+C 47 58 : R L L R R T T T T T T T L L L L L T T L
+C 49 78 : T T R T L L L L L T T T R R L F F T
+C 33 11 : L F R F T F F L R L L L L T L L F T
+C 7 24 : T F T F F R F L L L L R F F T T F F T
+C 7 18 : T T F F T L F L L L L T R R T L
+C 89 98 : L T R L L L L L
+C 95 97 : L R T L L L L T T T R L
+C 21 14 : R L L L L T T L L F R R T T T L T F F T
+C 85 81 : T F F F R T T L L L L L L L L L T R R R R T L L
+C 96 96 : L L L L
+C 1 23 : R L T T R L L L L T R T L F L L R T
+C 88 99 : R T T L L L L L
+C 50 62 : F F T T R L L L L
+C 9 38 : T L T T F L L L L T R T T R F R T T
+C 62 42 : F F L R F L T R R L L L L L L F R T L F T F T F R T F F
+C 28 19 : L R R T T T L L L L
+C 21 31 : F R L L L L R
+C 55 25 : F T R L L L L F F R T
+C 69 48 : T R R L L L L L L
+C 13 13 : T T F T L R L L L L L T T T R L
+C 13 23 : R L L L L
+C 37 37 : L R T L F F L L R R T L L L L L T R R R R R R F R F
+C 26 11 : L T T L T R L T F F R L L L L L L T L F
+C 56 56 : T F T L L F T T T R L L L L L F R T F T
+C 77 95 : T T T L F T L L L L L L
+C 70 70 : R F T R F F F T F F F T L L L L T T F F T F F R R T
+C 79 89 : R T T R L L L L T R R T T T T T L F T
+C 29 18 : T F T L F L T R F L L L L L T R R T
+C 72 91 : R R R T F T F L L L L L L F T R R
+C 77 83 : T L F R T L L L L T T L F T R F F F T
+C 37 18 : L R L R T L L L L L T R T L L T R R T L T R T
+C 33 32 : T T T R T T T L L L L R R T T R R
+C 90 91 : L T T R L L L L L T F T T
+C 46 43 : R R R L L L L L L T L
+C 18 39 : F T L L L L L R F F F F T F T T F F F F R
+C 84 93 : L T L R L L L L L L T T T R F F T T
+C 29 15 : L L L L R T L F F R L T T F T F T T T
+C 49 48 : T L T R R T L L L L L T T
+C 51 52 : F R L L L L L F T F R
+C 51 61 : T T T T F R L T T L L L L T F
+C 88 89 : F R T L L L T T L L L L L L R T L R T F T T T T T
+C 58 89 : L T L L L L F L T T T L F L
+C 1 31 : R R T L T T T L R T T F L L L L L T R L L L T
+C 29 29 : L L L L R T L F T F
+C 15 21 : T F L L L L T R L R T L F F T L T R F
+C 90 71 : L F R T L T R T L L L L L L T T F T R T T
+C 61 52 : F F L L L L L L F F T F
+C 12 12 : F R T L L L L L F F T F F T
+C 5 28 : F F F R F L L L L T T T T
+C 19 7 : R T L L L L T R R T T F L
+C 33 32 : L T T T T L L L L L F L R T T R T L T T F
+C 55 45 : L L L L T T F R T L R
+C 15 26 : F T T R L F L T L L L L R F
+C 96 96 : L L L L
+C 59 59 : L L L L
+C 96 85 : L L L L L L L T
+C 73 84 : T L L F L T L R L L L L T T L L T T F T R F
+C 90 91 : L F R T L L L L L L R
+C 85 96 : R L L L L L L T L T
+C 33 30 : L L L L L L T R R T T F F R T
+C 54 19 : L T T T F T R L L L L L F F R F R T R L R
+C 14 14 : T F L L L L T F
+C 98 89 : L T T T L T T L L L L L L T R T
+C 23 32 : F T F L L L L L
+C 24 23 : L L L L T T L R T L R L T
+C 49 59 : R R F T T T T T F L L L L L L T R R T
+C 33 22 : T T L L T T R T L L L L T L T R F T
+C 88 78 : F L T F T L L L L T R R F F T T T
+C 42 45 : T T T T F F F T T L T R L L L L L T R
+C 48 47 : R L L L L R R
+C 96 82 : L L L L F T F F F T F F L L F F F
+C 19 29 : T T R R T L L L L L L T R F T L R R R T T T T F T
+C 20 30 : R L L T L L L L L L T T F R F T F L R L R
+C 39 38 : L F L L L L T T L L L R R
+C 66 44 : T T T T L L F R R T L L L L
+C 49 49 : R R R T L L L L T R R T
+C 18 39 : R L L L L L R
+C 24 30 : T F T T L L L L F L R R T T T F T
+C 5 36 : F F F F T L L T L L L L T R R T T F L R
+C 37 18 : T T T T R T F R L T L L L L T R R R F T F T R T T
+C 10 32 : F F R F L L L L R T F L R T F L T L L L F T
+C 71 91 : T L L L L T R R R T L R
+C 70 82 : F T L L L L T R F T R
+C 42 66 : T T T T F F R L L L L T R T T T L F T R F R
+C 98 99 : L T T T L L L L L L F T R R T F L
+C 45 36 : T F F T F R L L L L L L F R T L T R
+C 8 18 : F T T T L F L L L L R T T L R T T T L R L
+C 42 56 : T T F F F R R T L L L L T L L R T R L T R T L T T L T
+C 41 64 : F T T F T T T T F R L L L L F
+C 59 48 : L L L L L L L L T F T L
+C 23 11 : T R R R F R T L R L L T T L L L L F F
+C 65 62 : T F F F T F L L L L T F T
+C 7 27 : T T F F T F T R L T L T R R T L L L L T T T F R T L L T T F
+C 25 21 : T F F F L L L L
+C 6 8 : F T T T T F R F L L L L L L L R T T L T R T R T
+C 81 70 : L L L L L L L L L L T T
+C 89 88 : L T F T L L L L T T F T T T F R T L L L L T R
+C 33 11 : L L L L L L T L L F
+C 55 44 : L F T R L L L L T T L F R T L R T R
+C 14 1 : T F L L L L F R
+C 74 94 : R T T L R L T T L L L L T F
+C 88 99 : R L L L L L
+C 24 25 : L L L L F F T
+C 24 24 : L L L L F T
+C 64 54 : F T T R T L F T T F T L L L L F F T R
+C 89 89 : T L L L L R T L L T
+C 14 16 : R T L L L L T T R F T T T T T T
+C 97 85 : T F T T F T T T F L L L L F T R L
+C 80 81 : L T L L L L T T T R R R T
+C 89 89 : L T T T F R T T T T T L L L L T R R R T T T F T
+C 91 71 : L R L T R R L L L L L L T R
+C 29 6 : L L L L L L T T T R T T T L R L F F T
+C 20 33 : R F T F T L L T L L L L L F
+C 72 70 : R T R T L R R L L L L T R R
+C 86 74 : R L T T T F T L L L L L F T T L F
+C 35 54 : R R L L L L L L T R T T R R R L T
+C 71 72 : T L L L L L T T T R R L L
+C 93 86 : F F F T T L L L L L F T T T
+C 33 25 : L L L L T F R T L L L L L F L R F
+C 31 13 : F L F T L F L L L L L L
+C 55 27 : L L L L F T F T L F F R F
+C 63 55 : F L F T L F R T F T R R L L L L T
+C 12 45 : R R T R L L L L T R L R R L R L F T T T F L
+C 86 85 : R R T L L L L T T F T R T R R
+C 25 13 : T R L L L L L
+C 78 49 : R T T T F T T F R L L L L L T R T L
+C 71 94 : R F T R F F L T L R L L L L T T T T T T T T T T T T L T F
+C 90 81 : L T T R L L L L T T R T
+C 63 62 : L L L L T T T
+C 50 63 : R T R F R L L L L T L T T T R T R R R T F L T R
+C 53 53 : R T L L L L
+C 99 98 : L L L L T T T
+C 63 53 : T R L T F T L L L L F T T T T F T F F
+C 15 13 : T F L L L L T T F L L L
+C 63 52 : L T L L R T L R R L L L L L L T L T R L T T T F T L
+C 21 21 : R L L L L R L T R T T L
+C 26 19 : T L T R F F F L L L L L
+C 18 35 : R R F F L L T L R L L L L F F T F F R T F R T
+C 23 22 : F T T L L T T L L L L L L F R T L F R T R

data/maze/100/train_C1_RWs_10M.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f61e18a319c19cd0fd195bb137bc80e4884e0aee1423fa7c24b2f30c9b121082
+size 2580000000

data/maze/100/train_C1_RWs_10M.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46e886b808c04989e1ef9b6bfc2b477d09260a6fd7bbf5179a4f678ce6129fb0
+size 1136936300

data/maze/100/train_C1_RWs_500K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30ba4e3b32009237da05444b5392cac253a9059e2dc5671975acba16334db7c3
+size 129000000

data/maze/100/train_C1_RWs_500K.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a03510e92cdfc04b1dbde08ef7e1b98060b92a56eea9e7a9247be49451fe62d
+size 56431106

data/maze/100/train_E1_RWs_500K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5676d7d588400d3bc8c5194ccaf22da87f23d04a53b00ff26ba69de332690bb8
+size 225000000

data/maze/100/train_E1_RWs_500K.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d41e1d6fc1b12545929b6e2e66acc1b864ad7ed5517b665a0eeb196e9aa4f686
+size 91127539

data/maze/100/train_H1_RWs_10M.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a974da897e8ba8c84bf9966cef2626eb09e850c1daade05a073f217de63941a4
+size 2580000000

data/maze/100/train_H1_RWs_10M.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d332814e333452c36e7e48fab8f71d5b3da8e2707c8291e5954d1d6671008364
+size 1135542770

data/maze/100/train_I1_RWs_10M.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eafb794d2654c3315ee14f89353fd1d2d46a92d9a72a413671a40ee6848e3d60
+size 2580000000

data/maze/100/train_I1_RWs_10M.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94106631193b1ff9b51d782d3e577e61f1a1cc313ca9746d7e4efa7c4b54464d
+size 1134435361

data/maze/100/val_A1_RWs_10K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5257317a14b2967f9e0e907dc65b4e917fb64436a9da555862c105e0bc0b85ae
+size 2580000

data/maze/100/val_A1_RWs_1K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f4aff60224a2f1095311df2cd1bbba705683a47abd385873a3c85f9a6158ef5
+size 258000

data/maze/100/val_C1_RWs_10K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a577be487e14d730b192bf251fa03d0ada4b285885bb9745dfd504d3ca1aca0e
+size 2580000

data/maze/100/val_E1_RWs_10K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1367fcd972f578a42f7d710f6e00fec6014b82e86daaec022c589fd14ba81c9
+size 4500000

data/maze/100/val_H1_RWs_10K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89ffe8c0b18b472430875c01856f4cabe1a4a1726d024c97bd6b51546148304e
+size 2580000

data/maze/100/val_I1_RWs_10K.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6100ea04ea792eb4c8140bbbe52b31a283f95842ff33b62758958a8081a2cef
+size 2580000

data/maze/create_maze.py

@@ -0,0 +1,447 @@
+import networkx as nx
+import random
+import os
+import argparse
+import numpy
+
+def generate_maze(n, edge_prob):
+    # Create a directed grid graph with random edge removal
+    G = nx.DiGraph()
+    for i in range(n*n):
+        G.add_node(i)
+    
+    # Add edges with probability edge_prob
+    for i in range(n):
+        for j in range(n):
+            node = i*n + j
+            # # up
+            # if i > 0 and random.random() < edge_prob:
+            #     G.add_edge(node, (i-1)*n + j)
+            #     G.add_edge((i-1)*n + j, node)
+            # down
+            if i < n-1 and random.random() < edge_prob:
+                G.add_edge(node, (i+1)*n + j)
+                G.add_edge((i+1)*n + j, node)
+            # # left
+            # if j > 0 and random.random() < edge_prob:
+            #     G.add_edge(node, i*n + j-1)
+            #     G.add_edge(i*n + j-1, node)
+            # right
+            if j < n-1 and random.random() < edge_prob:
+                G.add_edge(node, i*n + j+1)
+                G.add_edge(i*n + j+1, node)
+    
+    return G
+
+def print_grid(G, n, file=None):
+    def write_line(text):
+        if file is None:
+            print(text, end="")
+        else:
+            file.write(text)
+    
+    for i in range(n-1):
+        # Print row edges
+        for j in range(n):
+            write_line("+")
+            if j < n-1 and G.has_edge(i*n + j, i*n + j+1):
+                write_line("---")
+            elif j < n-1:
+                write_line("   ")
+        write_line("\n")
+        
+        # Print column edges
+        for j in range(n):
+            if G.has_edge(i*n + j, (i+1)*n + j):
+                write_line("|")
+            else:
+                write_line(" ")
+            if j < n-1:
+                write_line("   ")
+        write_line("\n")
+       
+    # Print bottom border
+    for j in range(n):
+        write_line("+")
+        if j < n-1 and G.has_edge((n-1)*n + j, (n-1)*n + j+1):
+            write_line("---")
+        elif j < n-1:
+            write_line("   ")
+    write_line("\n")
+
+def get_reachable_nodes(G, target_node):  
+    # Get the transitive closure of the graph  
+    TC = nx.transitive_closure(G)  
+    # Find the predecessors in the transitive closure (nodes that can reach the target_node)  
+    reachable_from = TC.predecessors(target_node)  
+    return list(reachable_from)
+
+def obtain_reachability():
+    reachability = {}  
+    pairs = 0
+    for node in maze_graph.nodes():  
+        reachability[node] = get_reachable_nodes(maze_graph, node)
+        pairs += len(reachability[node])
+    return reachability, pairs
+
+def random_walk(source_node, target_node):
+    stack = [source_node]
+    visited = [] # to eliminate cycles
+
+    while stack != []:
+        cur_node = stack.pop()
+        visited.append(cur_node)
+        if cur_node == target_node:
+            return visited
+        
+        adj = list(maze_graph.successors(cur_node))
+        anc = list(reachability[target_node])
+        anc.append(target_node)
+        
+        remaining = [element for element in adj if element in anc and element not in visited] #if we want the path to contain cycles, we should remove "and element not in visited"
+    
+        if len(remaining) == 0:
+            return random_walk(source_node, target_node) # for non-DAGs
+
+        next_node = random.choice(remaining)
+        stack.append(next_node)
+
+    return visited
+
+def seq2act(path):
+    actions = []
+    for i in range(1, len(path)):
+        diff = path[i] - path[i-1]
+        if diff == -n:
+            actions.append('N')
+        elif diff == n:
+            actions.append('S')
+        elif diff == -1:
+            actions.append('W')
+        elif diff == 1:
+            actions.append('E')
+    return actions
+
+
+def wall_directions(node):
+    """Return the list of NESW directions that hit a wall from `node`.
+
+    A direction is a "wall" when the adjacent in-grid cell exists but there is
+    no edge to it in the maze graph (i.e. the move is illegal).
+    """
+    i, j = divmod(node, n)
+    dirs = []
+    # North
+    if i > 0 and not maze_graph.has_edge(node, (i - 1) * n + j):
+        dirs.append('N')
+    # South
+    if i < n - 1 and not maze_graph.has_edge(node, (i + 1) * n + j):
+        dirs.append('S')
+    # West
+    if j > 0 and not maze_graph.has_edge(node, i * n + (j - 1)):
+        dirs.append('W')
+    # East
+    if j < n - 1 and not maze_graph.has_edge(node, i * n + (j + 1)):
+        dirs.append('E')
+    return dirs
+
+
+def corrupt_one_token(path):
+    """Turn a correct node path into a wrong (wall-hitting) action sequence.
+
+    Takes the correct actions of `path` and replaces ONE move (at a random
+    position, not necessarily the last) with an illegal direction that hits a
+    wall from that move's starting cell. The remaining tokens are kept as-is.
+    Returns the corrupted action list, or None if no position can be corrupted.
+    """
+    actions = seq2act(path)
+    if not actions:
+        return None
+    # actions[i] is the move taken from node path[i]; try positions in random
+    # order until we find one that has a wall to bump into.
+    idxs = list(range(len(actions)))
+    random.shuffle(idxs)
+    for i in idxs:
+        walls = wall_directions(path[i])  # illegal dirs from path[i]; excludes the legal actions[i]
+        if walls:
+            d = random.choice(walls)
+            return actions[:i] + [d] + actions[i + 1:]
+    return None
+
+
+def select_task(tasks_config, is_train=True):
+    """
+    Randomly select a task based on the configured percentages.
+    For training: uses 'train' percentage from tasks_config
+    For testing: uses 'test' percentage from tasks_config
+    Returns the selected task ID.
+    """
+    key = 'train' if is_train else 'test'
+    percentages = [(task_id, config[key]) for task_id, config in tasks_config.items()]
+    
+    # Generate random number between 0 and 100
+    rand = random.random() * 100
+    cumsum = 0
+    
+    for task_id, pct in percentages:
+        cumsum += pct
+        if rand < cumsum:
+            return task_id
+    
+    # Fallback to the last task if rounding errors occur
+    return percentages[-1][0]
+
+def create_dataset(i, tasks_config):
+    train_set = []
+    test_set = []
+    train_num_per_pair = max(i,1)
+    for target_node in range(num_nodes):
+        cnt = 0 # to avoid some target not appear in training dataset
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if (data[source_node][target_node] == 1):
+                if maze_graph.has_edge(source_node, target_node):
+                    task_id = select_task(tasks_config, is_train=True)
+                    if task_id == 'A':
+                        train_set.append(['A', source_node, target_node] + seq2act([source_node, target_node]))
+                    else:
+                        print(f"Error: Task {task_id} is not yet defined. Skipping training data generation for this entry.")
+                        
+                for ii in range(train_num_per_pair):
+                    task_id = select_task(tasks_config, is_train=True)
+                    if task_id == 'A':
+                        train_set.append(['A', source_node, target_node] + seq2act(random_walk(source_node, target_node)) )
+                    else:
+                        print(f"Error: Task {task_id} is not yet defined. Skipping training data generation for this entry.")
+                        
+            if (data[source_node][target_node] == -1):
+                task_id = select_task(tasks_config, is_train=False)
+                if task_id == 'A':
+                    test_set.append(['A', source_node, target_node] + seq2act(random_walk(source_node, target_node)))
+                else:
+                    print(f"Error: Task {task_id} is not yet defined. Skipping test data generation for this entry.")
+                    
+    return train_set, test_set
+
+def add_x(train_set, test_set, tasks_config):
+    cnt = 0
+    for target_node in range(num_nodes):
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node not in reachability[target_node]:
+                cnt += 1
+
+    prob_in_test = len(test_set) / cnt * 0.2
+    prob_in_train = min(len(train_set) / cnt * 0.2, 1 - prob_in_test)
+    train_repeat = max(int(len(train_set) / cnt * 0.15 / prob_in_train), 1)
+    print(prob_in_train, prob_in_test, train_repeat)
+
+    for target_node in range(num_nodes):
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node not in reachability[target_node]:
+                coin = random.random()
+                if coin < prob_in_train:
+                    for _ in range(train_repeat):
+                        task_id = select_task(tasks_config, is_train=True)
+                        if task_id == 'A':
+                            train_set.append(['A', source_node, target_node, 'x'])
+                        else:
+                            print(f"Error: Task {task_id} is not yet defined. Skipping training data generation for this entry.")
+                        
+                elif coin > 1 - prob_in_test:
+                    task_id = select_task(tasks_config, is_train=False)
+                    if task_id == 'A':
+                        test_set.append(['A', source_node, target_node, 'x'])
+                    else:
+                        print(f"Error: Task {task_id} is not yet defined. Skipping test data generation for this entry.")
+
+    return train_set, test_set
+
+
+def add_wrong_paths(train_set, test_set, tasks_config, wrong_ratio):
+    """Append wall-hitting wrong paths to the datasets.
+
+    Each wrong path is built from a CORRECT path (source -> target) by replacing
+    ONE move token (at a random position, not necessarily the last) with an
+    illegal direction that hits a wall, then appending 'x' at the end.
+    Format: ['A', source, target, <moves with one illegal move somewhere>, 'x'].
+    During training only the final 'x' token is supervised (see train_maze.py /
+    get_batch); every other token is masked out of the loss, so the model is
+    not taught to imitate the wrong trajectory and must inspect the whole
+    sequence to flag that a wall was hit.
+    """
+    if wrong_ratio <= 0:
+        return train_set, test_set
+
+    num_train_wrong = int(len(train_set) * wrong_ratio)
+    num_test_wrong = int(len(test_set) * wrong_ratio)
+
+    # Collect reachable (source, target) pairs, split like the correct data.
+    train_pairs = []
+    test_pairs = []
+    for target_node in range(num_nodes):
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node in reachability[target_node]:
+                if data[source_node][target_node] == 1:
+                    train_pairs.append((source_node, target_node))
+                elif data[source_node][target_node] == -1:
+                    test_pairs.append((source_node, target_node))
+
+    def gen_wrong(pairs, count, is_train):
+        out = []
+        if not pairs or count <= 0:
+            return out
+        attempts = 0
+        max_attempts = count * 50 + 100
+        while len(out) < count and attempts < max_attempts:
+            attempts += 1
+            source_node, target_node = random.choice(pairs)
+            path = random_walk(source_node, target_node)
+            if not path or len(path) < 2:
+                continue
+            corrupted = corrupt_one_token(path)
+            if corrupted is None:
+                continue
+            task_id = select_task(tasks_config, is_train=is_train)
+            if task_id != 'A':
+                continue
+            out.append(['A', source_node, target_node] + corrupted + ['x'])
+        return out
+
+    train_wrong = gen_wrong(train_pairs, num_train_wrong, True)
+    test_wrong = gen_wrong(test_pairs, num_test_wrong, False)
+    print(f'Added {len(train_wrong)} wrong (wall-hit) paths to train, {len(test_wrong)} to test.')
+    return train_set + train_wrong, test_set + test_wrong
+
+
+def obtain_stats(dataset):
+    max_len = 0
+    pairs = set()
+
+    for data in dataset:
+        max_len = max(max_len, len(data))
+        pairs.add((data[0],data[-1]))
+
+    len_stats = [0]*(max_len + 1) 
+    
+    for data in dataset:
+        length = len(data)
+        len_stats[length] += 1
+        
+    print('number of source target pairs:', len(pairs))
+    for ii in range(3, len(len_stats)):
+        print(f'There are {len_stats[ii]} paths with length {ii-3}')
+
+
+def format_data(data):
+    # Format: task_id source target [remaining_tokens]
+    return ' '.join(str(token) for token in data) + '\n'
+        
+def write_dataset(dataset, file_name):
+    with open(file_name, "w") as file:
+        for data in dataset:
+            file.write(format_data(data))
+
+
+def parse_tasks(tasks_str):
+    """
+    Parse task specification string into a dictionary.
+    Format: "TaskID:train_percent:test_percent,TaskID:train_percent:test_percent,..."
+    Example: "A:50:50,B:30:30,C:20:20"
+    Returns: {"A": {"train": 50, "test": 50}, "B": {"train": 30, "test": 30}, ...}
+    Validates that all training percentages sum to 100% and all test percentages sum to 100%.
+    """
+    tasks = {}
+    for task_spec in tasks_str.split(','):
+        parts = task_spec.strip().split(':')
+        if len(parts) != 3:
+            raise ValueError(f"Invalid task specification: {task_spec}. Expected format: TaskID:train_percent:test_percent")
+        task_id, train_pct, test_pct = parts[0].strip(), int(parts[1].strip()), int(parts[2].strip())
+        if task_id in tasks:
+            raise ValueError(f"Duplicate task ID: {task_id}")
+        tasks[task_id] = {"train": train_pct, "test": test_pct}
+    
+    # Validate that percentages sum to 100%
+    total_train = sum(config["train"] for config in tasks.values())
+    total_test = sum(config["test"] for config in tasks.values())
+    
+    if total_train != 100:
+        raise ValueError(f"Training task percentages must sum to 100%, but got {total_train}%")
+    if total_test != 100:
+        raise ValueError(f"Test task percentages must sum to 100%, but got {total_test}%")
+    
+    return tasks
+
+
+if __name__ == "__main__":  
+    parser = argparse.ArgumentParser(description='Generate a maze based on the given parameters.')  
+    parser.add_argument('--grid_size', type=int, default=10, help='Size of the grid (n x n)')  
+    parser.add_argument('--edge_prob', type=float, default=0.6, help='Probability to keep an edge in the grid graph')  
+    parser.add_argument('--chance_in_train', type=float, default=0.5, help='Chance of a pair being in the training set')  
+    parser.add_argument('--num_of_paths', type=int, default=20, help='Number of paths per pair nodes in training dataset')  
+    parser.add_argument('--wrong_ratio', type=float, default=0.0,
+                        help='Fraction of extra wall-hitting wrong paths to add, relative to the '
+                             'number of correct paths (e.g. 0.2 = +20%%). Each wrong path ends in '
+                             'an illegal move + "x"; only the "x" step is supervised during training. '
+                             'Default 0.0 (disabled).')  
+    # Multi-task specification: comma-separated task identifiers with their train/test percentages
+    # Format: "TaskID:train_percent:test_percent,TaskID:train_percent:test_percent,..."
+    # Example: "A:100:100" or "A:50:50,B:30:30,C:20:20"
+    # Default is Task A (pathfinding) with 100% in both train and test datasets
+    parser.add_argument('--tasks', type=str, default='A:100:100', 
+                        help='Task identifiers with percentages. Format: "TaskID:train_pct:test_pct,TaskID:train_pct:test_pct,..." (default: A:100:100)')  
+
+    args = parser.parse_args()  
+    
+    # Parse task specifications
+    tasks_config = parse_tasks(args.tasks)
+
+    n = args.grid_size
+    edge_prob = args.edge_prob
+    num_nodes = n * n
+    chance_in_train = args.chance_in_train
+    num_of_paths = args.num_of_paths
+
+    maze_graph = generate_maze(n, edge_prob)
+    
+    folder_name = os.path.join(os.path.dirname(__file__), f'{num_nodes}')
+    if not os.path.exists(folder_name):  
+        os.makedirs(folder_name)
+    
+    # Save grid visualization to file
+    grid_file_path = os.path.join(folder_name, f'maze_{n}_{edge_prob}_{num_of_paths}.txt')
+    with open(grid_file_path, 'w') as f:
+        print_grid(maze_graph, n, file=f)
+    
+    print_grid(maze_graph, n)
+    reachability, feasible_pairs = obtain_reachability()
+
+    data = numpy.zeros([num_nodes,num_nodes])
+    for target_node in range(num_nodes):
+        cnt = 0 # to avoid some target not appear in training dataset
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node in reachability[target_node]:
+                if (maze_graph.has_edge(source_node, target_node)) or random.random() < chance_in_train or cnt < 1:
+                    data[source_node][target_node] = 1
+                    cnt += 1 
+                else:
+                    data[source_node][target_node] = -1
+                
+    train_set, test_set = create_dataset(num_of_paths, tasks_config)
+
+    train_set, test_set = add_wrong_paths(train_set, test_set, tasks_config, args.wrong_ratio)
+
+    obtain_stats(train_set)
+    print('number of source target pairs:', len(test_set))
+
+    write_dataset(train_set, os.path.join(folder_name, f'train_{num_of_paths}.txt'))
+    write_dataset(test_set, os.path.join(folder_name, f'test.txt'))
+    nx.write_graphml(maze_graph, os.path.join(folder_name, f'maze_graph.graphml'))

data/maze/create_multitask_maze.py

@@ -0,0 +1,1204 @@
+import networkx as nx
+import random
+import os
+import sys
+import argparse
+import numpy
+import math
+from tqdm import tqdm
+
+# Ensure project root is importable when running this script directly
+sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..')))
+from cli_utils import parse_count, format_count, parse_task_distribution
+
+# Default: 10 lowercase letters for node labels (can be overridden via --num_labels)
+NODE_LABELS = ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j']
+
+
+def make_label_list(num_labels):
+    """Generate a list of label strings for the given count.
+
+    Up to 26: single lowercase letters (a-z).
+    Above 26: l0, l1, l2, ... (prefixed to avoid collision with directions/tasks).
+    """
+    if num_labels <= 26:
+        return [chr(ord('a') + i) for i in range(num_labels)]
+    else:
+        return [f'l{i}' for i in range(num_labels)]
+
+
+def generate_maze(n, edge_prob):
+    # Create a directed grid graph with random edge removal
+    G = nx.DiGraph()
+    for i in range(n * n):
+        # Assign a random label from NODE_LABELS to each node
+        label = random.choice(NODE_LABELS)
+        G.add_node(i, label=label)
+
+    # Add edges with probability edge_prob
+    for i in range(n):
+        for j in range(n):
+            node = i * n + j
+            # # up
+            # if i > 0 and random.random() < edge_prob:
+            #     G.add_edge(node, (i-1)*n + j)
+            #     G.add_edge((i-1)*n + j, node)
+            # down
+            if i < n - 1 and random.random() < edge_prob:
+                G.add_edge(node, (i + 1) * n + j)
+                G.add_edge((i + 1) * n + j, node)
+            # # left
+            # if j > 0 and random.random() < edge_prob:
+            #     G.add_edge(node, i*n + j-1)
+            #     G.add_edge(i*n + j-1, node)
+            # right
+            if j < n - 1 and random.random() < edge_prob:
+                G.add_edge(node, i * n + j + 1)
+                G.add_edge(i * n + j + 1, node)
+
+    return G
+
+
+def print_grid(G, n, file=None):
+    def write_line(text):
+        if file is None:
+            print(text, end="")
+        else:
+            file.write(text)
+
+    for i in range(n - 1):
+        # Print row edges
+        for j in range(n):
+            write_line("+")
+            if j < n - 1 and G.has_edge(i * n + j, i * n + j + 1):
+                write_line("---")
+            elif j < n - 1:
+                write_line("   ")
+        write_line("\n")
+
+        # Print column edges and node labels (shifted one position to the right)
+        # First, print the edge and leading space
+        for j in range(n):
+            if G.has_edge(i * n + j, (i + 1) * n + j):
+                write_line("|")
+            else:
+                write_line(" ")
+
+            # Print label with spacing
+            # Print label of node (i, j) in the current cell (shifted right in the output)
+            label = G.nodes[i * n + j]['label']
+            write_line(f"{label}  ")
+        write_line("\n")
+
+    # Print bottom border
+    for j in range(n):
+        write_line("+")
+        if j < n - 1 and G.has_edge((n - 1) * n + j, (n - 1) * n + j + 1):
+            write_line("---")
+        elif j < n - 1:
+            write_line("   ")
+    write_line("\n")
+
+    # Print labels for the last row nodes
+    for j in range(n):
+        # Print label of node (n-1, j) below the border
+        label = G.nodes[(n - 1) * n + j]['label']
+        write_line(f" {label}  ")
+    write_line("\n")
+
+
+def get_reachable_nodes(G, target_node):
+    # Get the transitive closure of the graph
+    TC = nx.transitive_closure(G)
+    # Find the predecessors in the transitive closure (nodes that can reach the target_node)
+    reachable_from = TC.predecessors(target_node)
+    return list(reachable_from)
+
+
+def obtain_reachability():
+    reachability = {}
+    pairs = 0
+    for node in maze_graph.nodes():
+        reachability[node] = get_reachable_nodes(maze_graph, node)
+        pairs += len(reachability[node])
+    return reachability, pairs
+
+
+def random_walk(source_node, target_node, allow_cycles=False):
+    """Generate a random walk from source_node to target_node.
+
+    Args:
+        source_node: Starting node
+        target_node: Target node
+        allow_cycles: If False (default), path is acyclic. If True, path can contain cycles.
+
+    Returns:
+        List of nodes in the path, or empty list if no path found
+    """
+    stack = [source_node]
+    visited = []  # to track visited nodes for acyclic constraint
+
+    while stack != []:
+        cur_node = stack.pop()
+        visited.append(cur_node)
+        if cur_node == target_node:
+            return visited
+
+        adj = list(maze_graph.successors(cur_node))
+        anc = list(reachability[target_node])
+        anc.append(target_node)
+
+        if allow_cycles:
+            # Allow cycles: only check if node is reachable to target
+            remaining = [element for element in adj if element in anc]
+        else:
+            # Acyclic: check both reachability and no previous visits
+            remaining = [element for element in adj if element in anc and element not in visited]
+
+        if len(remaining) == 0:
+            return []  # no path found from this start/target
+
+        next_node = random.choice(remaining)
+        stack.append(next_node)
+
+    return visited
+
+
+def random_walk_ss(source_node, num_steps):
+    """Generate a single-source random walk of a fixed number of steps.
+
+    Args:
+        source_node: Starting node
+        num_steps: Number of steps to take
+
+    Returns:
+        List of nodes in the path
+    """
+    path = [source_node]
+    current = source_node
+    for _ in range(num_steps):
+        neighbors = list(maze_graph.successors(current))
+        if not neighbors:
+            break
+        current = random.choice(neighbors)
+        path.append(current)
+    return path
+
+
+def seq2act(path):
+    actions = []
+    for i in range(1, len(path)):
+        diff = path[i] - path[i - 1]
+        if diff == -n:
+            actions.append('N')
+        elif diff == n:
+            actions.append('S')
+        elif diff == -1:
+            actions.append('W')
+        elif diff == 1:
+            actions.append('E')
+    return actions
+
+
+def seq2turn(path, start_orientation='E'):
+    """Convert an absolute direction path to relative turns.
+
+    Each output token both turns and advances one step:
+    - F: keep facing direction and move forward
+    - L: turn left then move
+    - R: turn right then move
+    - T: turn around then move
+    """
+
+    absolute_actions = seq2act(path)
+    turns = []
+
+    left_of = {'N': 'W', 'W': 'S', 'S': 'E', 'E': 'N'}
+    right_of = {v: k for k, v in left_of.items()}
+    opposite_of = {'N': 'S', 'S': 'N', 'E': 'W', 'W': 'E'}
+
+    orientation = start_orientation
+    for action in absolute_actions:
+        if action == orientation:
+            turns.append('F')
+        elif action == left_of[orientation]:
+            turns.append('L')
+        elif action == right_of[orientation]:
+            turns.append('R')
+        elif action == opposite_of[orientation]:
+            turns.append('T')
+        else:
+            # Should not happen if the grid uses only NESW moves
+            continue
+        orientation = action
+
+    return turns
+
+
+def random_walk_with_cycles(start_node, walk_length):
+    path = [start_node]
+    current = start_node
+    for _ in range(walk_length):
+        neighbors = list(maze_graph.successors(current))
+        if not neighbors:
+            break
+        current = random.choice(neighbors)
+        path.append(current)
+    return path
+
+
+def shortest_path_to_label(source_node, target_label):
+    best_path = None
+    best_len = None
+    best_target = None
+    for node_id, attrs in maze_graph.nodes(data=True):
+        if attrs.get('label') != target_label:
+            continue
+        if node_id == source_node:
+            continue
+        if source_node not in reachability.get(node_id, []):
+            continue
+        try:
+            path = nx.shortest_path(maze_graph, source_node, node_id)
+        except nx.NetworkXNoPath:
+            continue
+        path_len = len(path)
+        if best_len is None or path_len < best_len or (path_len == best_len and node_id < best_target):
+            best_len = path_len
+            best_path = path
+            best_target = node_id
+    return best_path
+
+
+def select_task(tasks_config, is_train=True):
+    """
+    Randomly select a task based on the configured percentages.
+    For training: uses 'train' percentage from tasks_config
+    For testing: uses 'test' percentage from tasks_config
+    Returns the selected task ID.
+    """
+    key = 'train' if is_train else 'test'
+    percentages = [(task_id, config[key]) for task_id, config in tasks_config.items()]
+
+    # Generate random number between 0 and 100
+    rand = random.random() * 100
+    cumsum = 0
+
+    for task_id, pct in percentages:
+        cumsum += pct
+        if rand < cumsum:
+            return task_id
+
+    # Fallback to the last task if rounding errors occur
+    return percentages[-1][0]
+
+
+def _generate_one_entry(tasks_config, is_train):
+    """Generate a single multitask data entry. Returns None on failure or unknown task."""
+    task_id = select_task(tasks_config, is_train=is_train)
+    if task_id == 'A':
+        return create_data_entry_taskA(is_train=is_train)
+    elif task_id == 'B':
+        return create_data_entry_taskB(is_train=is_train)
+    elif task_id == 'C':
+        return create_data_entry_taskC(is_train=is_train)
+    elif task_id == 'D':
+        return create_data_entry_taskD(is_train=is_train)
+    elif task_id == 'E':
+        return create_data_entry_taskE(is_train=is_train)
+    elif task_id == 'F':
+        return create_data_entry_taskF(is_train=is_train)
+    elif task_id == 'G':
+        return create_data_entry_taskG(is_train=is_train)
+    elif task_id == 'H':
+        return create_data_entry_taskH(is_train=is_train)
+    elif task_id == 'I':
+        return create_data_entry_taskI(is_train=is_train)
+    else:
+        print(f"Warning: Unknown task ID '{task_id}'. Skipping this entry.")
+        return None
+
+
+def _worker_init(seed_base):
+    """Reseed RNGs in each worker so processes produce independent streams."""
+    pid = os.getpid()
+    random.seed(seed_base + pid)
+    numpy.random.seed((seed_base + pid) % (2 ** 31))
+
+
+def _worker_generate_batch(args_tuple):
+    """Worker entry point: generate `count` entries in this process."""
+    count, tasks_config, is_train = args_tuple
+    out = []
+    for _ in range(count):
+        entry = _generate_one_entry(tasks_config, is_train)
+        if entry is not None:
+            out.append(entry)
+    return out
+
+
+def create_multitask_dataset(num_of_data, tasks_config, is_train=True):
+    """
+    Generate a multitask dataset with multiple task types.
+
+    Args:
+        num_of_data: Number of data entries to generate
+        tasks_config: Dictionary with task configurations and percentages
+        is_train: If True, generate training data; if False, generate test data
+
+    Returns:
+        dataset: List of data entries for the specified tasks
+    """
+    # Parallel path: split work across processes (fork inherits maze_graph etc.)
+    nw = globals().get('num_workers', 1)
+    desc = f"Generating {'train' if is_train else 'test'} data"
+    if nw and nw > 1 and num_of_data > 0:
+        import multiprocessing as mp
+        import time
+        # Use many small batches so the progress bar updates smoothly.
+        batch_size = max(1, min(1000, math.ceil(num_of_data / (nw * 100))))
+        batches = []
+        remaining = num_of_data
+        while remaining > 0:
+            c = min(batch_size, remaining)
+            batches.append((c, tasks_config, is_train))
+            remaining -= c
+        seed_base = int(time.time()) + (0 if is_train else 10 ** 6)
+        ctx = mp.get_context('fork')
+        dataset = []
+        with ctx.Pool(processes=nw,
+                      initializer=_worker_init,
+                      initargs=(seed_base,)) as pool:
+            with tqdm(total=num_of_data, desc=desc) as pbar:
+                for r in pool.imap_unordered(_worker_generate_batch, batches):
+                    dataset.extend(r)
+                    pbar.update(len(r))
+        return dataset
+
+    # Serial path (original behavior)
+    dataset = []
+    for _ in tqdm(range(num_of_data), desc=desc):
+        data_entry = _generate_one_entry(tasks_config, is_train)
+        if data_entry is not None:
+            dataset.append(data_entry)
+    return dataset
+
+
+def create_data_entry_taskA(is_train=True):
+    """
+    Generate one Task A (pathfinding) data entry.
+    Randomly samples a reachable source-target pair that belongs to the requested split
+    and returns: ['A', source_node, target_node, <actions...>].
+    """
+    while True:
+        if path_type_tag == 'RWs':
+            # Single Source (Taxi routing style): Pick a start, walk random steps, find end
+            source_node = random.randrange(num_nodes)
+            num_steps = random.randint(5, num_nodes)
+            path = random_walk_ss(source_node, num_steps)
+            if not path or len(path) < 2:
+                continue
+            target_node = path[-1]
+        else:
+            # Goal-oriented: Pick start and end first
+            source_node = random.randrange(num_nodes)
+            target_node = random.randrange(num_nodes)
+
+            if source_node == target_node:
+                continue
+
+            # Ensure the target is reachable from the source to avoid dead walks
+            if source_node not in reachability.get(target_node, []):
+                continue
+
+        label = data[source_node][target_node]
+        if is_train and label != 1:
+            continue
+        if (not is_train) and label != -1:
+            continue
+
+        if path_type_tag != 'RWs':
+            path = random_walk(source_node, target_node, allow_cycles=allow_cycles)
+
+        if not path:
+            continue
+        actions = seq2act(path)
+
+        # Optional minimum path length filter
+        if min_path_len > 0 and len(actions) < min_path_len:
+            continue
+
+        # for both training and test data, needs to return full format with answer part, this is needed to generate validation data during training
+        return ['A', source_node, target_node, ':'] + actions
+
+
+def create_data_entry_taskC(is_train=True):
+    """Generate one Task C entry (turn-based pathfinding).
+
+    Uses the same reachable-pair sampling as Task A, but encodes the
+    path with relative turns assuming the agent starts facing East.
+    Output format: ['C', source_node, target_node, ':', <turns...>] by default.
+    When cl_mode is True, append the node label after every L/R turn.
+    """
+
+    while True:
+        if path_type_tag == 'RWs':
+            # Single Source logic for Task C
+            source_node = random.randrange(num_nodes)
+            num_steps = random.randint(5, num_nodes)
+            path = random_walk_ss(source_node, num_steps)
+            if not path or len(path) < 2:
+                continue
+            target_node = path[-1]
+        else:
+            source_node = random.randrange(num_nodes)
+            target_node = random.randrange(num_nodes)
+
+            if source_node == target_node:
+                continue
+
+            if source_node not in reachability.get(target_node, []):
+                continue
+
+        label = data[source_node][target_node]
+        if is_train and label != 1:
+            continue
+        if (not is_train) and label != -1:
+            continue
+
+        if path_type_tag != 'RWs':
+            path = random_walk(source_node, target_node, allow_cycles=allow_cycles)
+
+        if not path:
+            continue
+
+        turn_actions = seq2turn(path, start_orientation='E')
+
+        tokens = ['C', source_node, target_node, ':']
+        for step_idx, turn in enumerate(turn_actions):
+            tokens.append(turn)
+            if cl_mode and turn in ['L', 'R']:
+                node_id = path[step_idx]
+                tokens.append(maze_graph.nodes[node_id]['label'])
+
+        return tokens
+
+
+def create_data_entry_taskD(is_train=True):
+    """Generate one Task D entry (pathfinding to a target label).
+
+    Input provides a source node and a target label. The answer is a shortest
+    path (directions) to the nearest node with that label.
+    Format: ['D', source_node, target_label, ':', <directions...>]
+    """
+
+    while True:
+        source_node = random.randrange(num_nodes)
+        target_label = random.choice(NODE_LABELS)
+
+        path = shortest_path_to_label(source_node, target_label)
+        if not path:
+            continue
+
+        actions = seq2act(path)
+        if not actions:
+            continue
+
+        return ['D', source_node, target_label, ':'] + actions
+
+
+def create_data_entry_taskE(is_train=True):
+    """
+    Task E (pathfinding with label observations).
+    - Only split segments when the direction changes (i.e., at turns).
+    - For each straight segment, let end_label be the label at the last node of this segment
+      (i.e., the turning node label, or the final node label if path ends).
+    - In that segment, keep ONLY the positions whose label == end_label.
+      Emit (dir, end_label) once for each kept position (so you may repeat the same pair).
+    Format: ['E', source_node, target_node, ':', dir1, lab1, dir2, lab2, ...]
+    """
+
+    while True:
+        if path_type_tag == 'RWs':
+            # Single Source (Taxi routing style): Pick a start, walk random steps, find end
+            source_node = random.randrange(num_nodes)
+            num_steps = random.randint(5, num_nodes)
+            path = random_walk_ss(source_node, num_steps)
+            if not path or len(path) < 2:
+                continue
+            target_node = path[-1]
+        else:
+            # Goal-oriented: Pick start and end first
+            source_node = random.randrange(num_nodes)
+            target_node = random.randrange(num_nodes)
+
+            if source_node == target_node:
+                continue
+
+            if source_node not in reachability.get(target_node, []):
+                continue
+
+            path = random_walk(source_node, target_node)
+            if not path or len(path) < 2:
+                continue
+
+        y = data[source_node][target_node]
+        if is_train and y != 1:
+            continue
+        if (not is_train) and y != -1:
+            continue
+
+        actions = seq2act(path)  # len(actions) == len(path)-1
+        if not actions:
+            continue
+
+        # Optional minimum path length filter
+        if min_path_len > 0 and len(actions) < min_path_len:
+            continue
+
+        # ---- 1) split only by turns (direction changes) ----
+        tokens = ['E', source_node, target_node, ':']
+
+        run_dir = actions[0]
+        run_labels = []  # labels of nodes visited during this run (node after each step)
+
+        for step_idx, direction in enumerate(actions):
+            node_id = path[step_idx + 1]
+            lab = maze_graph.nodes[node_id]['label']
+
+            # direction changed => flush previous run, then start new
+            if direction != run_dir:
+                # flush old run
+                end_label = run_labels[-1]
+                cnt = sum(1 for x in run_labels if x == end_label)
+                for _ in range(cnt):
+                    tokens.append(run_dir)
+                    tokens.append(end_label)
+
+                # start new run
+                run_dir = direction
+                run_labels = [lab]
+            else:
+                # still same direction => accumulate
+                run_labels.append(lab)
+
+        # ---- 2) flush last run ----
+        end_label = run_labels[-1]
+        cnt = sum(1 for x in run_labels if x == end_label)
+        for _ in range(cnt):
+            tokens.append(run_dir)
+            tokens.append(end_label)
+
+        return tokens
+
+
+def create_data_entry_taskF(is_train=True):
+    """Generate one Task F entry (label-based target identification).
+
+    Format: ['F', start_label, <directions...>, ':', target_label]
+    The start node is implicit: any node with the given start_label is valid.
+    """
+
+    start_label = random.choice(NODE_LABELS)
+    candidates = [node for node, attrs in maze_graph.nodes(data=True) if attrs.get('label') == start_label]
+    if not candidates:
+        return None
+
+    start_node = random.choice(candidates)
+    max_walk_len = max(1, 4 * n)
+    walk_length = random.randint(1, max_walk_len)
+
+    path = random_walk_with_cycles(start_node, walk_length)
+    end_node = path[-1]
+    directions = seq2act(path)
+
+    target_label = maze_graph.nodes[end_node]['label']
+
+    return ['F', start_label] + directions + [':', target_label]
+
+
+def create_data_entry_taskG(is_train=True):
+    """Generate one Task G entry (reachability choice with path as CoT).
+
+    Format: ['G', s1, s2, t1, t2, ':', source, target, <directions...>]
+    Exactly one of (s1->t1) or (s2->t2) is reachable.
+    """
+
+    while True:
+        source1 = random.randrange(num_nodes)
+        source2 = random.randrange(num_nodes)
+        target1 = random.randrange(num_nodes)
+        target2 = random.randrange(num_nodes)
+
+        if source1 == target1 or source2 == target2:
+            continue
+
+        reachable1 = source1 in reachability.get(target1, [])
+        reachable2 = source2 in reachability.get(target2, [])
+
+        if reachable1 == reachable2:
+            continue
+
+        if reachable1:
+            source_node, target_node = source1, target1
+        else:
+            source_node, target_node = source2, target2
+
+        path = random_walk(source_node, target_node)
+        if not path or len(path) < 2:
+            continue
+
+        actions = seq2act(path)
+        return ['G', source1, source2, target1, target2, ':', source_node, target_node] + actions
+
+
+def create_data_entry_taskH(is_train=True):
+    """Task H: Relative clockwise-index path encoding.
+
+    The walker starts at source facing East. At each step, feasible edges
+    are enumerated clockwise starting from the first direction after the
+    current facing direction. The output is the 1-based index of the
+    chosen direction among the feasible edges.
+
+    Format: ['H', source, target, ':', idx1, idx2, ...]
+    where each idx is a string '1'-'4'.
+    """
+    # Clockwise scan: starting from the current facing direction
+    CLOCKWISE_SCAN = {
+        'N': ['N', 'E', 'S', 'W'],
+        'E': ['E', 'S', 'W', 'N'],
+        'S': ['S', 'W', 'N', 'E'],
+        'W': ['W', 'N', 'E', 'S'],
+    }
+    DELTA = {'N': -n, 'S': n, 'E': 1, 'W': -1}
+
+    while True:
+        if path_type_tag == 'RWs':
+            source_node = random.randrange(num_nodes)
+            num_steps = random.randint(5, num_nodes)
+            path = random_walk_ss(source_node, num_steps)
+            if not path or len(path) < 2:
+                continue
+            target_node = path[-1]
+        else:
+            source_node = random.randrange(num_nodes)
+            target_node = random.randrange(num_nodes)
+            if source_node == target_node:
+                continue
+            if source_node not in reachability.get(target_node, []):
+                continue
+
+        label = data[source_node][target_node]
+        if is_train and label != 1:
+            continue
+        if (not is_train) and label != -1:
+            continue
+
+        if path_type_tag != 'RWs':
+            path = random_walk(source_node, target_node, allow_cycles=allow_cycles)
+
+        if not path or len(path) < 2:
+            continue
+        actions = seq2act(path)
+        if not actions:
+            continue
+
+        if min_path_len > 0 and len(actions) < min_path_len:
+            continue
+
+        # Convert absolute directions to relative clockwise indices
+        facing = 'E'
+        tokens = ['H', source_node, target_node, ':']
+        valid = True
+
+        for step_idx, action in enumerate(actions):
+            current_node = path[step_idx]
+            scan_order = CLOCKWISE_SCAN[facing]
+            feasible = []
+            for d in scan_order:
+                neighbor = current_node + DELTA[d]
+                if 0 <= neighbor < num_nodes and maze_graph.has_edge(current_node, neighbor):
+                    feasible.append(d)
+
+            if action not in feasible:
+                valid = False
+                break
+
+            idx = feasible.index(action) + 1  # 1-based
+            tokens.append(str(idx))
+            facing = action  # update facing to direction actually moved
+
+        if not valid:
+            continue
+
+        return tokens
+
+
+def create_data_entry_taskI(is_train=True):
+    """Task I: Absolute clockwise-index path encoding (fixed North reference).
+
+    Like Task H, but feasible edges are always enumerated clockwise starting
+    from a FIXED North reference (N, E, S, W) regardless of the direction the
+    walker just moved. The walker therefore does NOT track a facing direction:
+    its state is the current node alone. The output is the 1-based index of the
+    chosen direction among the node's feasible edges in this fixed N->E->S->W
+    order.
+
+    This isolates "state-conditioned retrieval" (must read the node's feasible
+    edge set to emit/decode the index) from "facing tracking": compared to
+    Task A it adds only retrieval; compared to Task H it removes facing tracking.
+
+    Format: ['I', source, target, ':', idx1, idx2, ...]
+    where each idx is a string '1'-'4'.
+    """
+    # Fixed clockwise scan order from North (no dependence on facing)
+    FIXED_SCAN = ['N', 'E', 'S', 'W']
+    DELTA = {'N': -n, 'S': n, 'E': 1, 'W': -1}
+
+    while True:
+        if path_type_tag == 'RWs':
+            source_node = random.randrange(num_nodes)
+            num_steps = random.randint(5, num_nodes)
+            path = random_walk_ss(source_node, num_steps)
+            if not path or len(path) < 2:
+                continue
+            target_node = path[-1]
+        else:
+            source_node = random.randrange(num_nodes)
+            target_node = random.randrange(num_nodes)
+            if source_node == target_node:
+                continue
+            if source_node not in reachability.get(target_node, []):
+                continue
+
+        label = data[source_node][target_node]
+        if is_train and label != 1:
+            continue
+        if (not is_train) and label != -1:
+            continue
+
+        if path_type_tag != 'RWs':
+            path = random_walk(source_node, target_node, allow_cycles=allow_cycles)
+
+        if not path or len(path) < 2:
+            continue
+        actions = seq2act(path)
+        if not actions:
+            continue
+
+        if min_path_len > 0 and len(actions) < min_path_len:
+            continue
+
+        # Convert absolute directions to fixed-North clockwise indices
+        tokens = ['I', source_node, target_node, ':']
+        valid = True
+
+        for step_idx, action in enumerate(actions):
+            current_node = path[step_idx]
+            feasible = []
+            for d in FIXED_SCAN:
+                neighbor = current_node + DELTA[d]
+                if 0 <= neighbor < num_nodes and maze_graph.has_edge(current_node, neighbor):
+                    feasible.append(d)
+
+            if action not in feasible:
+                valid = False
+                break
+
+            idx = feasible.index(action) + 1  # 1-based
+            tokens.append(str(idx))
+            # NOTE: no facing update — reference stays fixed at North.
+
+        if not valid:
+            continue
+
+        return tokens
+
+
+# Add 'x' entries for unreachable pairs, only used if we allow unreachable pairs in the dataset, and currently it is not used.
+def add_x(train_set, test_set, tasks_config):
+    cnt = 0
+    for target_node in range(num_nodes):
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node not in reachability[target_node]:
+                cnt += 1
+
+    prob_in_test = len(test_set) / cnt * 0.2
+    prob_in_train = min(len(train_set) / cnt * 0.2, 1 - prob_in_test)
+    train_repeat = max(int(len(train_set) / cnt * 0.15 / prob_in_train), 1)
+    print(prob_in_train, prob_in_test, train_repeat)
+
+    for target_node in range(num_nodes):
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node not in reachability[target_node]:
+                coin = random.random()
+                if coin < prob_in_train:
+                    for _ in range(train_repeat):
+                        task_id = select_task(tasks_config, is_train=True)
+                        if task_id == 'A':
+                            train_set.append(['A', source_node, target_node, 'x'])
+                        else:
+                            print(
+                                f"Error: Task {task_id} is not yet defined. Skipping training data generation for this entry.")
+
+                elif coin > 1 - prob_in_test:
+                    task_id = select_task(tasks_config, is_train=False)
+                    if task_id == 'A':
+                        test_set.append(['A', source_node, target_node, 'x'])
+                    else:
+                        print(
+                            f"Error: Task {task_id} is not yet defined. Skipping test data generation for this entry.")
+
+    return train_set, test_set
+
+
+def create_data_entry_taskB(is_train=True):
+    """Generate one Task B entry (target identification).
+
+    Training format: B <start> <directions...> : <end_label> <E> <S> <W> <N>
+    Test format:     B <start> <directions...>
+    Walk length is random up to 4 * grid_size (n).
+    """
+
+    def get_neighbor_labels(node):
+        neighbors_order = [(1, 'E'), (n, 'S'), (-1, 'W'), (-n, 'N')]
+        labels = []
+        for offset, _ in neighbors_order:
+            neighbor_id = node + offset
+            if maze_graph.has_edge(node, neighbor_id):
+                labels.append(maze_graph.nodes[neighbor_id]['label'])
+            else:
+                labels.append('/')
+        return labels
+
+    start_node = random.randint(0, num_nodes - 1)
+    max_walk_len = max(1, 4 * n)
+    walk_length = random.randint(1, max_walk_len)
+
+    path = random_walk_with_cycles(start_node, walk_length)
+    end_node = path[-1]
+    directions = seq2act(path)
+
+    end_label = maze_graph.nodes[end_node]['label']
+    neighbor_labels = get_neighbor_labels(end_node)
+
+    # for both training and test data, return answer without target node id (only label + neighbors)
+    return ['B', start_node] + directions + [':', end_label] + neighbor_labels
+
+
+def obtain_stats(dataset, is_train=True):
+    """
+    Compute and print statistics for a multitask dataset.
+
+    Args:
+        dataset: List of data entries (can be mixed tasks)
+        is_train: If True, label output as training data; else as test data
+    """
+    dataset_type = "Training" if is_train else "Test"
+    print(f'\n{dataset_type} Dataset Statistics:')
+    print('=' * 80)
+
+    # Separate entries by task
+    taskA_entries = [entry for entry in dataset if entry[0] == 'A']
+    taskB_entries = [entry for entry in dataset if entry[0] == 'B']
+    taskC_entries = [entry for entry in dataset if entry[0] == 'C']
+    taskD_entries = [entry for entry in dataset if entry[0] == 'D']
+    taskE_entries = [entry for entry in dataset if entry[0] == 'E']
+    taskF_entries = [entry for entry in dataset if entry[0] == 'F']
+    taskG_entries = [entry for entry in dataset if entry[0] == 'G']
+    taskH_entries = [entry for entry in dataset if entry[0] == 'H']
+    taskI_entries = [entry for entry in dataset if entry[0] == 'I']
+
+    print(f'Total entries: {len(dataset)}')
+    print(f'  Task A entries: {len(taskA_entries)}')
+    print(f'  Task B entries: {len(taskB_entries)}')
+    print(f'  Task C entries: {len(taskC_entries)}')
+    print(f'  Task D entries: {len(taskD_entries)}')
+    print(f'  Task E entries: {len(taskE_entries)}')
+    print(f'  Task F entries: {len(taskF_entries)}')
+    print(f'  Task G entries: {len(taskG_entries)}')
+    print(f'  Task H entries: {len(taskH_entries)}')
+    print(f'  Task I entries: {len(taskI_entries)}')
+
+    # Task A statistics
+    if taskA_entries:
+        print(f'\nTask A (Path finding):')
+        # Extract source-target pairs from task A entries
+        # Format: A source target directions...
+        pairs = set()
+        for entry in taskA_entries:
+            if len(entry) >= 3:
+                source = entry[1]
+                target = entry[2]
+                pairs.add((source, target))
+
+        num_pairs = len(pairs)
+        num_entries = len(taskA_entries)
+        avg_per_pair = num_entries / num_pairs if num_pairs > 0 else 0
+
+        print(f'  Number of source-target pairs: {num_pairs}')
+        print(f'  Number of data entries: {num_entries}')
+        print(f'  Average entries per pair: {avg_per_pair:.2f}')
+
+    # Task B statistics
+    if taskB_entries:
+        print(f'\nTask B (Target identification):')
+        # Extract source nodes from task B entries
+        # Format: B source directions...
+        source_nodes = set()
+        for entry in taskB_entries:
+            if len(entry) >= 2:
+                source = entry[1]
+                source_nodes.add(source)
+
+        num_sources = len(source_nodes)
+        num_entries = len(taskB_entries)
+        avg_per_source = num_entries / num_sources if num_sources > 0 else 0
+
+        print(f'  Number of source nodes: {num_sources}')
+        print(f'  Number of data entries: {num_entries}')
+        print(f'  Average entries per source node: {avg_per_source:.2f}')
+
+    # Task C statistics (mirrors Task A pair counting)
+    if taskC_entries:
+        print(f'\nTask C (Turn-based path finding):')
+        pairs = set()
+        for entry in taskC_entries:
+            if len(entry) >= 3:
+                source = entry[1]
+                target = entry[2]
+                pairs.add((source, target))
+        num_pairs = len(pairs)
+        num_entries = len(taskC_entries)
+        avg_per_pair = num_entries / num_pairs if num_pairs > 0 else 0
+        print(f'  Number of source-target pairs: {num_pairs}')
+        print(f'  Number of data entries: {num_entries}')
+        print(f'  Average entries per pair: {avg_per_pair:.2f}')
+
+    if taskD_entries:
+        print(f'\nTask D (Path finding to label):')
+        print(f'  Number of data entries: {len(taskD_entries)}')
+
+    if taskE_entries:
+        print(f'\nTask E (Path finding with labels):')
+        print(f'  Number of data entries: {len(taskE_entries)}')
+
+    if taskF_entries:
+        print(f'\nTask F (Target label identification):')
+        print(f'  Number of data entries: {len(taskF_entries)}')
+
+    if taskG_entries:
+        print(f'\nTask G (Reachability choice):')
+        print(f'  Number of data entries: {len(taskG_entries)}')
+
+    if taskH_entries:
+        print(f'\nTask H (Relative clockwise-index path):')
+        print(f'  Number of data entries: {len(taskH_entries)}')
+
+
+def format_data(data, no_task_tag=False):
+    # Format: task_id source target [remaining_tokens]
+    # If no_task_tag is True, remove the first token (task identifier)
+    if no_task_tag and len(data) > 0:
+        # Remove the task identifier (first token)
+        return ' '.join(str(token) for token in data[1:]) + '\n'
+    else:
+        return ' '.join(str(token) for token in data) + '\n'
+
+
+def write_dataset(dataset, file_name, no_task_tag=False):
+    with open(file_name, "w") as file:
+        for data in dataset:
+            file.write(format_data(data, no_task_tag))
+
+
+def parse_tasks(tasks_str):
+    """
+    Parse task specification string into a dictionary.
+    Format: "A<ratio>B<ratio>C<ratio>..." where ratios determine portions.
+    Example: "A1" (100% A), "A1B1" (50% A, 50% B), "A3B2" (60% A, 40% B), "A2B1C1" (50% A, 25% B, 25% C)
+    Returns: {"A": {"train": 50, "test": 50}, "B": {"train": 50, "test": 50}, ...}
+    Test data follows the same ratio as training data.
+    """
+    weights = parse_task_distribution(tasks_str, default_task='A')
+    total_ratio = sum(weights.values())
+    tasks_config = {}
+    for task_id, ratio in weights.items():
+        percentage = (ratio / total_ratio) * 100
+        tasks_config[task_id] = {"train": percentage, "test": percentage}
+    return tasks_config
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Generate a maze based on the given parameters.')
+    parser.add_argument('--grid_size', type=int, default=10, help='Size of the grid (n x n)')
+    parser.add_argument('--edge_prob', type=float, default=0.6, help='Probability to keep an edge in the grid graph')
+    parser.add_argument('--chance_in_train', type=float, default=0.5, help='Chance of a pair being in the training set')
+    parser.add_argument('--num_train_dataset', type=parse_count, default='10M',
+                        help='Number of training data entries to generate (supports K/M/B, default: 50000)')
+    parser.add_argument('--num_test_dataset', type=parse_count, default=10000,
+                        help='Number of test data entries to generate (supports K/M/B, default: 10000)')
+    # Multi-task specification:
+    parser.add_argument('--tasks', type=str, default='H1',
+                        help='Task identifiers with ratios. Format: "A<ratio>B<ratio>C<ratio>...". Examples: "A1" (100%% A), "A1B1" (50%% A, 50%% B), "A3B2" (60%% A, 40%% B), "A1D1F1" (mix A/D/F). Default: A1')
+    parser.add_argument('--CL', action=argparse.BooleanOptionalAction, default=False,
+                        help='Enable Task C label mode (append node labels after L/R turns) and add _CL_ in filenames')
+    parser.add_argument('--graph_file', type=str, default=None,
+                        help='Optional path to an existing GraphML file; if provided, skip random graph generation and use this graph instead.')
+    parser.add_argument('--path_type', type=str, default='RWs', choices=['RWc', 'RWa', 'RWs'],
+                        help='Path generation type: RWc (random walk with cycles), RWa (random walk acyclic, default), RWs (single source random walk). "shortest" is not implemented yet.')
+    # Arguments for task tag handling
+    parser.add_argument('--no_task_tag', action='store_true', default=False,
+                        help='Remove task identifiers from generated data. When enabled, output files will have _NT suffix and data entries start directly with node numbers.')
+    parser.add_argument('--both', action='store_true', default=False,
+                        help='Generate both versions (with and without task tags). When set, --no_task_tag is ignored and two datasets are produced.')
+    parser.add_argument('--num_labels', type=int, default=10,
+                        help='Number of distinct node labels (default: 10). Up to 26 uses a-z; above 26 uses l0, l1, ...')
+    parser.add_argument('--min_path_len', type=int, default=0,
+                        help='Minimum raw path length (in steps) for generated entries. 0 means no minimum (default).')
+    parser.add_argument('--num_workers', type=int, default=256,
+                        help='Number of parallel worker processes for dataset generation (default: 1 = serial). Uses fork; requires Linux/macOS.')
+
+    args = parser.parse_args()
+
+    # Override NODE_LABELS if --num_labels is specified
+    if args.num_labels != 10:
+        NODE_LABELS = make_label_list(args.num_labels)
+        print(f"Using {args.num_labels} labels: {NODE_LABELS[:5]}{'...' if args.num_labels > 5 else ''}")
+    min_path_len = args.min_path_len
+    num_workers = args.num_workers
+
+    # Parse task specifications
+    tasks_config = parse_tasks(args.tasks)
+    tasks_str = args.tasks  # Keep the original tasks string for filenames
+    tasks_tag = f"{tasks_str}_CL" if args.CL else tasks_str
+    cl_mode = args.CL
+    no_task_tag = args.no_task_tag  # Get the no_task_tag flag
+
+    # Parse path_type for filenames (RWc = cyclic, RWa = acyclic, RWs = single source)
+    allow_cycles = (args.path_type == 'RWc')
+    path_type_tag = args.path_type
+    tasks_tag = f"{tasks_tag}_{path_type_tag}"
+
+    # Include num_labels and min_path_len in tags when non-default
+    if args.num_labels != 10:
+        tasks_tag = f"{tasks_tag}_L{args.num_labels}"
+    if args.min_path_len > 0:
+        tasks_tag = f"{tasks_tag}_P{args.min_path_len}"
+
+    edge_prob = args.edge_prob
+    chance_in_train = args.chance_in_train
+    num_train_dataset = args.num_train_dataset
+    num_test_dataset = args.num_test_dataset
+    train_label = format_count(num_train_dataset)
+    test_label = format_count(num_test_dataset)
+    graph_file = args.graph_file
+
+    if graph_file:
+        graph_path = graph_file
+        if not os.path.isabs(graph_file):
+            default_dir = os.path.join(os.path.dirname(__file__), f'{args.grid_size * args.grid_size}')
+            candidate = os.path.join(default_dir, graph_file)
+            if os.path.exists(candidate):
+                graph_path = candidate
+        print(f"Loading maze graph from {graph_path}...")
+        maze_graph = nx.read_graphml(graph_path)
+        # Ensure node ids are integers (GraphML loader returns strings)
+        try:
+            int_map = {node: int(node) for node in maze_graph.nodes()}
+            maze_graph = nx.relabel_nodes(maze_graph, int_map, copy=True)
+        except ValueError:
+            pass
+
+        num_nodes = len(maze_graph.nodes)
+        n = int(math.isqrt(num_nodes))
+        if n * n != num_nodes:
+            print(
+                f"Warning: provided graph has {num_nodes} nodes; not a perfect square grid. Proceeding with derived size {n}.")
+    else:
+        n = args.grid_size
+        num_nodes = n * n
+        maze_graph = generate_maze(n, edge_prob)
+
+    folder_name = os.path.join(os.path.dirname(__file__), f'{num_nodes}')
+    if not os.path.exists(folder_name):
+        os.makedirs(folder_name)
+
+    # Always save grid visualization to file with consistent naming convention
+    maze_viz_tag = f"{tasks_str}_CL" if cl_mode else tasks_str
+    maze_viz_tag = f"{maze_viz_tag}_{path_type_tag}"
+    if args.num_labels != 10:
+        maze_viz_tag = f"{maze_viz_tag}_L{args.num_labels}"
+    # Graph and visualization files: always save without _NT, regardless of no_task_tag/both
+    # This ensures they are compatible with both data formats
+    grid_file_path = os.path.join(folder_name, f'maze_{maze_viz_tag}_{n}_{edge_prob}.txt')
+    with open(grid_file_path, 'w') as f:
+        print_grid(maze_graph, n, file=f)
+
+    # Always print visualization to the screen
+    print_grid(maze_graph, n)
+
+    reachability, feasible_pairs = obtain_reachability()
+
+    # This is for generating pairs for training and test datasets for task A
+    data = numpy.zeros([num_nodes, num_nodes])
+    for target_node in range(num_nodes):
+        cnt = 0  # to avoid some target not appear in training dataset
+        for source_node in range(num_nodes):
+            if source_node == target_node:
+                continue
+            if source_node in reachability[target_node]:
+                if (maze_graph.has_edge(source_node, target_node)) or random.random() < chance_in_train or cnt < 1:
+                    data[source_node][target_node] = 1
+                    cnt += 1
+                else:
+                    data[source_node][target_node] = -1
+
+    train_set = create_multitask_dataset(num_train_dataset, tasks_config, is_train=True)
+    test_set = create_multitask_dataset(num_test_dataset, tasks_config, is_train=False)
+
+    obtain_stats(train_set, is_train=True)
+    obtain_stats(test_set, is_train=False)
+
+    # Build graph tag for output files (without _NT)
+    graph_tag = f"{tasks_str}_CL" if cl_mode else tasks_str
+    graph_tag = f"{graph_tag}_{path_type_tag}"
+    # Include num_labels in graph tag when non-default (different label sets = different graphs)
+    if args.num_labels != 10:
+        graph_tag = f"{graph_tag}_L{args.num_labels}"
+
+    # Generate datasets based on both/no_task_tag flags
+    if args.both:
+        # Generate with task tags (no _NT suffix)
+        tag_with_tags = tasks_tag
+        write_dataset(train_set, os.path.join(folder_name, f'train_{tag_with_tags}_{train_label}.txt'),
+                      no_task_tag=False)
+        write_dataset(test_set, os.path.join(folder_name, f'test_{tag_with_tags}_{test_label}.txt'), no_task_tag=False)
+
+        # Generate without task tags (_NT suffix)
+        tag_without_tags = f"{tasks_tag}_NT"
+        write_dataset(train_set, os.path.join(folder_name, f'train_{tag_without_tags}_{train_label}.txt'),
+                      no_task_tag=True)
+        write_dataset(test_set, os.path.join(folder_name, f'test_{tag_without_tags}_{test_label}.txt'),
+                      no_task_tag=True)
+
+        # Save graph files: both without _NT and with _NT
+        nx.write_graphml(maze_graph, os.path.join(folder_name, f'maze_graph_{graph_tag}.graphml'))
+        nx.write_graphml(maze_graph, os.path.join(folder_name, f'maze_graph_{graph_tag}_NT.graphml'))
+
+        print("Generated both with-tag and without-tag datasets and graph files.")
+    else:
+        # Original logic: generate only one version based on no_task_tag
+        output_tasks_tag = tasks_tag
+        if no_task_tag:
+            output_tasks_tag = f"{tasks_tag}_NT"
+        write_dataset(train_set, os.path.join(folder_name, f'train_{output_tasks_tag}_{train_label}.txt'), no_task_tag)
+        write_dataset(test_set, os.path.join(folder_name, f'test_{output_tasks_tag}_{test_label}.txt'), no_task_tag)
+
+        # Save graph file with appropriate suffix
+        graph_suffix = "_NT" if no_task_tag else ""
+        nx.write_graphml(maze_graph, os.path.join(folder_name, f'maze_graph_{graph_tag}{graph_suffix}.graphml'))

data/maze/prepare_minigpt.py

@@ -0,0 +1,164 @@
+import os
+import pickle
+import numpy as np
+import re
+import argparse
+
+parser = argparse.ArgumentParser(description='Create the dataset based on the given parameters.')  
+parser.add_argument('--num_nodes', type=int, default=100, help='Number of nodes in the graph')  
+parser.add_argument('--num_of_paths', type=int, default=20, help='Number of paths per pair nodes in training dataset')  
+args = parser.parse_args()  
+
+num_nodes = args.num_nodes
+
+if(args.num_of_paths == 0):
+    train_file_path = os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train.txt')
+    val_file_path = os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/test.txt')
+else:
+    train_file_path = os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{args.num_of_paths}.txt')
+    val_file_path = os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/test.txt')
+# test_file_path = os.path.join(os.path.dirname(__file__), 'test.txt')
+
+with open(train_file_path, 'r') as f:
+    train_data = f.read()
+print(f"length of train dataset in characters: {len(train_data):,}")
+
+with open(val_file_path, 'r') as f:
+    val_data = f.read()
+print(f"length of val dataset in characters: {len(val_data):,}")
+
+all_data = train_data + val_data
+
+def find_characters(data_string):
+    pattern = r'\d+|\D'
+    matches = re.findall(pattern, data_string)
+    return set(matches)
+
+def process_reasoning(s):
+    split_text = s.split('\n')
+    #split_text = [s + '\n' for s in split_text if s != ""]
+    ret = []
+    for st in split_text:
+        if(st != ""):
+            enc_str = encode(st) + [1]
+            ret += enc_str +[0] * (block_size + 1 - len(enc_str))
+    return ret
+
+def get_block_size(s):
+    split_text = s.split('\n')
+    #split_text = [s + '\n' for s in split_text if s != ""]
+    ret = []
+    bs = 0
+    for st in split_text:
+        if(st != ""):
+            enc_str = encode(st) + [1]
+            bs = max(bs, len(enc_str))
+    return bs
+
+
+def encode_string(s, stonum):
+    ss = s.split(" ")
+    encoded_string = [stonum[ch] for ch in ss]
+    return encoded_string
+
+def decode_string(l, numtos):
+    dec = ""
+    for i in l:
+        dec = dec + numtos[i] + " "
+    return dec[:-1]
+
+
+# get all the unique characters that occur in this text
+chars = sorted(list(find_characters(all_data)))
+# direction tokens for maze paths
+direction_tokens = ['N','S','E','W']
+# task tokens for multi-task support
+task_tokens = ['A', 'B', 'C', 'D', 'E', 'F', 'G']
+# special tokens: 'x' marks a wall-hit / unreachable (wrong-path) terminator
+special_tokens = ['x']
+# vocab = node ids + PAD + newline + direction tokens + task tokens + special tokens
+vocab_size = num_nodes + 2 + len(direction_tokens) + len(task_tokens) + len(special_tokens)
+print("all the unique characters:", ' '.join(chars))
+print(f"vocab size: {vocab_size:,}")
+
+# create a mapping from characters to integers
+stoi = {}
+itos = {}
+
+for i in range(num_nodes):
+    stoi[str(i)] = i+2
+    itos[i+2] = str(i)
+
+# map direction tokens after the node id tokens
+base = 2 + num_nodes
+for idx, tok in enumerate(direction_tokens):
+    stoi[tok] = base + idx
+    itos[base + idx] = tok
+
+# map task tokens after direction tokens
+base = 2 + num_nodes + len(direction_tokens)
+for idx, tok in enumerate(task_tokens):
+    stoi[tok] = base + idx
+    itos[base + idx] = tok
+
+# map special tokens (e.g. 'x') after task tokens
+base = 2 + num_nodes + len(direction_tokens) + len(task_tokens)
+for idx, tok in enumerate(special_tokens):
+    stoi[tok] = base + idx
+    itos[base + idx] = tok
+
+stoi['[PAD]'] = 0
+itos[0] = '[PAD]'
+stoi['\n'] = 1
+itos[1] = '\n'
+
+def encode(s):
+    return encode_string(s, stoi) # encoder: take a string, output a list of integers
+def decode(l):
+    return decode_string(l, itos) # decoder: take a list of integers, output a string
+
+# encode both to integers
+block_size = (max(get_block_size(train_data), get_block_size(val_data)) // 32 + 1) * 32
+
+print(f"the block size is {block_size}")
+
+train_ids = process_reasoning(train_data)
+
+val_ids = process_reasoning(val_data)
+
+print(f"train has {len(train_ids):,} tokens")
+print(f"val has {len(val_ids):,} tokens")
+
+# export to bin files
+train_ids = np.array(train_ids, dtype=np.uint16)
+val_ids = np.array(val_ids, dtype=np.uint16)
+
+if(args.num_of_paths == 0):
+    train_ids.tofile(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train.bin'))
+    val_ids.tofile(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/val.bin'))
+else:
+    train_ids.tofile(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{args.num_of_paths}.bin'))
+    val_ids.tofile(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/val.bin'))
+
+
+unreachable = False; simple_format = True
+if 'x' in chars:
+    unreachable = True
+if ':' in chars:
+    simple_format = False
+    
+
+# save the meta information as well, to help us encode/decode later
+meta = {
+    'unreachable': unreachable,
+    'simple_format': simple_format,
+    'block_size': block_size,
+    'vocab_size': vocab_size,
+    'itos': itos,
+    'stoi': stoi,
+}
+
+print(stoi)
+print(itos)
+with open(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/meta.pkl'), 'wb') as f:
+    pickle.dump(meta, f)

data/maze/prepare_multitask_minigpt.py

@@ -0,0 +1,323 @@
+import os
+import sys
+import pickle
+import numpy as np
+import re
+import argparse
+from tqdm import tqdm
+
+# Ensure project root is importable when running this script directly
+sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..')))
+from cli_utils import parse_count, format_count
+
+parser = argparse.ArgumentParser(description='Create the multitask dataset based on the given parameters.')
+parser.add_argument('--num_nodes', type=int, default=100, help='Number of nodes in the graph')
+parser.add_argument('--num_train_dataset', type=parse_count, default='10M',
+                    help='Number of training data entries to use (supports K/M/B, default: 50000)')
+parser.add_argument('--num_test_dataset', type=parse_count, default=10000,
+                    help='Number of test data entries to use (supports K/M/B, default: 10000)')
+parser.add_argument('--tasks', type=str, default='H1',
+                    help='Task specification (e.g., A1, A1B1, A3B2, A1D1F1). Default: A1')
+parser.add_argument('--CL', action=argparse.BooleanOptionalAction, default=False,
+                    help='Enable Task C label mode (append node labels after L/R turns) and add _CL_ in filenames')
+parser.add_argument('--path_type', type=str, default='RWs', choices=['RWc', 'RWa', 'RWs'],
+                    help='Path generation type: RWc (random walk with cycles), RWa (random walk acyclic, default), RWs (single source random walk).')
+# Arguments for task tag handling
+parser.add_argument('--no_task_tag', action='store_true', default=False,
+                    help='Data files do not contain task identifiers (A, B, C, etc.). When enabled, task tokens will not be included in vocabulary and data parsing will skip task tags.')
+parser.add_argument('--both', action='store_true', default=False,
+                    help='Process both versions (with and without task tags). When set, --no_task_tag is ignored and two sets of bin files and meta files are produced.')
+parser.add_argument('--num_labels', type=int, default=10,
+                    help='Number of distinct node labels (default: 10). Must match the value used in data generation.')
+parser.add_argument('--num_workers', type=int, default=1,
+                    help='Number of parallel worker processes for encoding (default: 1 = serial). Uses fork; requires Linux/macOS.')
+args = parser.parse_args()
+
+num_nodes = args.num_nodes
+tasks_str = args.tasks
+tasks_tag_base = f"{tasks_str}_CL" if args.CL else tasks_str
+# Add path type tag for filenames
+path_type_tag = args.path_type
+tasks_tag_base = f"{tasks_tag_base}_{path_type_tag}"
+# Include num_labels in tag when non-default (match create_multitask_maze.py)
+if args.num_labels != 10:
+    tasks_tag_base = f"{tasks_tag_base}_L{args.num_labels}"
+
+train_label = format_count(args.num_train_dataset)
+test_label = format_count(args.num_test_dataset)
+num_labels = args.num_labels
+
+def first_existing(paths):
+    for p in paths:
+        if os.path.exists(p):
+            return p
+    return paths[0]
+
+
+def process_data_for_tag_mode(no_task_tag_mode, tasks_tag_suffix=""):
+    """Process data for a specific task tag mode."""
+    # Construct tasks_tag for this mode
+    if no_task_tag_mode:
+        tasks_tag = f"{tasks_tag_base}_NT"
+        if tasks_tag_suffix:
+            tasks_tag = f"{tasks_tag}_{tasks_tag_suffix}"
+    else:
+        tasks_tag = tasks_tag_base
+        if tasks_tag_suffix:
+            tasks_tag = f"{tasks_tag}_{tasks_tag_suffix}"
+
+    # Find input files
+    train_file_path = first_existing([
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{tasks_tag}_{train_label}.txt'),
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{tasks_tag}_{args.num_train_dataset}.txt'),
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{tasks_str}_{train_label}.txt'),
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{tasks_str}_{args.num_train_dataset}.txt'),
+    ])
+    val_file_path = first_existing([
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/test_{tasks_tag}_{test_label}.txt'),
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/test_{tasks_tag}_{args.num_test_dataset}.txt'),
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/test_{tasks_str}_{test_label}.txt'),
+        os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/test_{tasks_str}_{args.num_test_dataset}.txt'),
+    ])
+
+    print(f"\nProcessing mode: {'Without task tags' if no_task_tag_mode else 'With task tags'}")
+    print(f"Training file: {train_file_path}")
+    print(f"Test file: {val_file_path}")
+
+    with open(train_file_path, 'r') as f:
+        train_data = f.read()
+    print(f"length of train dataset in characters: {len(train_data):,}")
+
+    with open(val_file_path, 'r') as f:
+        val_data = f.read()
+    print(f"length of val dataset in characters: {len(val_data):,}")
+
+    all_data = train_data + val_data
+
+    chars = sorted(list(find_characters(all_data)))
+    direction_tokens = ['N', 'S', 'E', 'W', 'L', 'R', 'F', 'T']
+    # Only include task tokens if no_task_tag_mode is False
+    task_tokens = [] if no_task_tag_mode else ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I']
+    label_tokens = ([chr(ord('a') + i) for i in range(num_labels)] if num_labels <= 26
+                    else [f'l{i}' for i in range(num_labels)])
+    label_tokens.append('/')  # separator token for neighbor labels
+    special_tokens = [':']
+    # Adjust vocab_size calculation
+    vocab_size = num_nodes + 2 + len(direction_tokens) + len(task_tokens) + len(label_tokens) + len(special_tokens)
+    print("all the unique characters:", ' '.join(chars))
+    print(f"vocab size: {vocab_size:,}")
+    print(f"No task tag mode: {'Enabled' if no_task_tag_mode else 'Disabled'}")
+
+    stoi = {}
+    itos = {}
+
+    for i in range(num_nodes):
+        stoi[str(i)] = i + 2
+        itos[i + 2] = str(i)
+
+    base = 2 + num_nodes
+    for idx, tok in enumerate(direction_tokens):
+        stoi[tok] = base + idx
+        itos[base + idx] = tok
+
+    # Only add task tokens to vocabulary if no_task_tag_mode is False
+    if not no_task_tag_mode:
+        base = 2 + num_nodes + len(direction_tokens)
+        for idx, tok in enumerate(task_tokens):
+            stoi[tok] = base + idx
+            itos[base + idx] = tok
+        base = 2 + num_nodes + len(direction_tokens) + len(task_tokens)
+    else:
+        base = 2 + num_nodes + len(direction_tokens)
+
+    for idx, tok in enumerate(label_tokens):
+        stoi[tok] = base + idx
+        itos[base + idx] = tok
+
+    base = base + len(label_tokens)
+    for idx, tok in enumerate(special_tokens):
+        stoi[tok] = base + idx
+        itos[base + idx] = tok
+
+    stoi['[PAD]'] = 0
+    itos[0] = '[PAD]'
+    stoi['\n'] = 1
+    itos[1] = '\n'
+
+    def encode(s):
+        ss = s.split(" ")
+        return [stoi[ch] for ch in ss]
+
+    def decode(l):
+        return ' '.join(itos[i] for i in l)
+
+    # Calculate block_size with theoretical minimum
+    n = int(num_nodes ** 0.5)  # grid size
+    if no_task_tag_mode:
+        theoretical_min_tokens = num_nodes + 3
+    else:
+        theoretical_min_tokens = num_nodes + 4
+
+    theoretical_min_block_size = (theoretical_min_tokens // 32 + 1) * 32
+
+    nw = args.num_workers
+
+    def get_block_size(s, desc="scan block size"):
+        split_text = s.split('\n')
+        if nw and nw > 1 and len(split_text) > 0:
+            import multiprocessing as mp
+            chunk_size = max(1, min(20000, len(split_text) // (nw * 50) or 1))
+            chunks = _chunk_list(split_text, chunk_size)
+            ctx = mp.get_context('fork')
+            bs = 0
+            with ctx.Pool(processes=nw) as pool:
+                with tqdm(total=len(split_text), desc=desc) as pbar:
+                    for r in pool.imap_unordered(_prep_max_len_batch, chunks):
+                        if r > bs:
+                            bs = r
+                        pbar.update(chunk_size if pbar.n + chunk_size <= len(split_text)
+                                    else len(split_text) - pbar.n)
+            return bs
+        # Serial
+        bs = 0
+        for st in tqdm(split_text, desc=desc):
+            if st != "":
+                enc_str = encode(st) + [1]
+                bs = max(bs, len(enc_str))
+        return bs
+
+    data_block_size = (max(get_block_size(train_data, desc="scan train block size"),
+                           get_block_size(val_data, desc="scan val block size")) // 32 + 1) * 32
+    block_size = max(theoretical_min_block_size, data_block_size)
+    print(
+        f"the block size is {block_size} (theoretical min: {theoretical_min_block_size}, data-based: {data_block_size})")
+
+    def process_reasoning(s, desc="encode"):
+        split_text = s.split('\n')
+        if nw and nw > 1 and len(split_text) > 0:
+            import multiprocessing as mp
+            chunk_size = max(1, min(10000, len(split_text) // (nw * 100) or 1))
+            chunks = _chunk_list(split_text, chunk_size)
+            ctx = mp.get_context('fork')
+            ret = []
+            with ctx.Pool(processes=nw,
+                          initializer=_prep_worker_init,
+                          initargs=(stoi, block_size)) as pool:
+                with tqdm(total=len(split_text), desc=desc) as pbar:
+                    # Use imap (ordered) to preserve original line order in output.
+                    for i, r in enumerate(pool.imap(_prep_encode_batch, chunks)):
+                        ret.extend(r)
+                        step = len(chunks[i])
+                        pbar.update(step)
+            return ret
+        # Serial
+        ret = []
+        for st in tqdm(split_text, desc=desc):
+            if st != "":
+                enc_str = encode(st) + [1]
+                ret += enc_str + [0] * (block_size + 1 - len(enc_str))
+        return ret
+
+    train_ids = process_reasoning(train_data, desc="encode train")
+    val_ids = process_reasoning(val_data, desc="encode val")
+
+    print(f"train has {len(train_ids):,} tokens")
+    print(f"val has {len(val_ids):,} tokens")
+
+    train_ids = np.array(train_ids, dtype=np.uint16)
+    val_ids = np.array(val_ids, dtype=np.uint16)
+
+    # Save bin files with appropriate tag
+    train_ids.tofile(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/train_{tasks_tag}_{train_label}.bin'))
+    val_ids.tofile(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/val_{tasks_tag}_{test_label}.bin'))
+
+    unreachable = 'x' in chars
+    simple_format = ':' not in chars
+
+    meta = {
+        'unreachable': unreachable,
+        'simple_format': simple_format,
+        'block_size': block_size,
+        'vocab_size': vocab_size,
+        'itos': itos,
+        'stoi': stoi,
+        'no_task_tag': no_task_tag_mode,
+    }
+
+    print(stoi)
+    print(itos)
+    with open(os.path.join(os.path.dirname(__file__), f'{args.num_nodes}/meta_{tasks_tag}.pkl'), 'wb') as f:
+        pickle.dump(meta, f)
+
+    print(f"Saved files with tag: {tasks_tag}")
+    return tasks_tag
+
+
+def find_characters(data_string):
+    pattern = r'\d+|\D'
+    matches = re.findall(pattern, data_string)
+    return set(matches)
+
+
+# ---- Parallel worker helpers (must be at module scope for pickling) ----
+_W_STOI = None
+_W_BLOCK_SIZE = None
+
+
+def _prep_worker_init(stoi_arg, block_size_arg):
+    global _W_STOI, _W_BLOCK_SIZE
+    _W_STOI = stoi_arg
+    _W_BLOCK_SIZE = block_size_arg
+
+
+def _prep_max_len_batch(lines):
+    """Return the max encoded length (tokens + EOL) over a batch of lines."""
+    bs = 0
+    for st in lines:
+        if st == "":
+            continue
+        # encoded length = number of space-separated tokens + 1 (EOL token)
+        L = st.count(" ") + 2
+        if L > bs:
+            bs = L
+    return bs
+
+
+def _prep_encode_batch(lines):
+    """Encode + pad a batch of lines; returns a flat list of token IDs."""
+    stoi = _W_STOI
+    bs1 = _W_BLOCK_SIZE + 1
+    out = []
+    for st in lines:
+        if st == "":
+            continue
+        enc = [stoi[ch] for ch in st.split(" ")]
+        enc.append(1)
+        out.extend(enc)
+        out.extend([0] * (bs1 - len(enc)))
+    return out
+
+
+def _chunk_list(lst, chunk_size):
+    return [lst[i:i + chunk_size] for i in range(0, len(lst), chunk_size)]
+
+
+# Main execution
+if args.both:
+    # Process both versions
+    print("=" * 60)
+    print("Generating both with-tag and without-tag versions")
+    print("=" * 60)
+
+    # Process with task tags
+    process_data_for_tag_mode(no_task_tag_mode=False)
+
+    # Process without task tags
+    process_data_for_tag_mode(no_task_tag_mode=True)
+
+    print("=" * 60)
+    print("Successfully generated both with-tag and without-tag datasets.")
+    print("=" * 60)
+else:
+    # Original logic: generate only one version based on no_task_tag
+    process_data_for_tag_mode(no_task_tag_mode=args.no_task_tag)