spatial_extractor.py

"""
RDE Spatial Field Extractor v1.2 (Auto-detect CSV columns)
Extracts 2D spatial arrays from OpenFOAM case directories for ML training.
Handles both scalar fields (p, T, H2, O2) and vector fields (U).
Auto-detects CSV column names for case matching.

Usage:
    python spatial_extractor.py --cases_dir ./ --cases RDE_01 RDE_02 ... RDE_12

Output:
    - spatial_fields.npz: (n_cases, n_timesteps, n_channels, H, W)
"""

import numpy as np
import json
import os
import re
import argparse
from pathlib import Path
import gzip
import pandas as pd


def sniff_csv_columns(csv_path):
    """
    Deep-dive: Auto-detect case identifier column and parameter columns.
    Returns mapping of standard names to actual CSV column names.
    """
    df = pd.read_csv(csv_path)
    cols = list(df.columns)
    
    print(f"  CSV columns found: {cols}")
    
    mapping = {}
    
    # Case ID detection (deep dive: check multiple patterns)
    case_patterns = ['case_id', 'case', 'Case', 'CASE', 'case_name', 'simulation', 'run']
    for pat in case_patterns:
        if pat in cols:
            mapping['case_id'] = pat
            break
    
    # If no explicit case column, check if first column looks like case names
    if 'case_id' not in mapping:
        first_col = cols[0]
        sample_vals = df[first_col].astype(str).tolist()
        # Check if values look like RDE_01, case_1, etc.
        if any(re.match(r'.*[_-]\d+', str(v)) for v in sample_vals[:5]):
            mapping['case_id'] = first_col
            print(f"  Using first column '{first_col}' as case identifier")
    
    # Parameter detection
    param_patterns = {
        'phi': ['phi', 'Phi', 'PHI', 'equivalence_ratio', 'ER'],
        'p0_pa': ['p0', 'p0_pa', 'P0', 'pressure', 'inlet_pressure', 'p_init'],
        'T0_k': ['T0', 'T0_k', 'temperature', 'inlet_temperature', 'T_init'],
        'cj_speed_ms': ['cj_speed', 'cantera_cj_speed_ms', 'CJ_speed', 'dcj', 'D_CJ']
    }
    
    for std_name, patterns in param_patterns.items():
        for pat in patterns:
            if pat in cols:
                mapping[std_name] = pat
                break
    
    print(f"  Column mapping: {mapping}")
    return mapping, df


def read_openfoam_header(field_path):
    """Read OpenFOAM file header to determine field type and dimensions."""
    field_path = Path(field_path)
    
    if not field_path.exists():
        gz_path = field_path.with_suffix(field_path.suffix + '.gz')
        if gz_path.exists():
            field_path = gz_path
        else:
            return {'is_uniform': False, 'is_vector': False}
    
    opener = gzip.open if str(field_path).endswith('.gz') else open
    
    with opener(field_path, 'rt', errors='replace') as f:
        header_lines = []
        for i, line in enumerate(f):
            header_lines.append(line)
            if i > 50:
                break
            if 'internalField' in line:
                break
    
    header_text = ''.join(header_lines)
    
    is_uniform = 'uniform' in header_text.lower() and 'nonuniform' not in header_text.lower()
    is_vector = 'volVectorField' in header_text
    is_scalar = 'volScalarField' in header_text
    
    if not is_vector and not is_scalar:
        for line in header_lines:
            stripped = line.strip()
            if stripped.startswith('(') and len(stripped.split()) > 2:
                is_vector = True
                break
    
    return {'is_uniform': is_uniform, 'is_vector': is_vector}


def read_openfoam_field(field_path, shape=(150, 300)):
    """
    Read OpenFOAM field file and reshape to 2D.
    Returns: (H, W) for scalar or (H, W, 3) for vector
    """
    field_path = Path(field_path)
    
    if not field_path.exists():
        gz_path = field_path.with_suffix(field_path.suffix + '.gz')
        if gz_path.exists():
            field_path = gz_path
        else:
            raise FileNotFoundError(f"Field file not found: {field_path}")
    
    field_info = read_openfoam_header(field_path)
    opener = gzip.open if str(field_path).endswith('.gz') else open
    
    with opener(field_path, 'rt', errors='replace') as f:
        lines = f.readlines()
    
    data_start = None
    n_cells = None
    
    for i, line in enumerate(lines):
        if 'internalField' in line:
            if 'uniform' in line.lower():
                val_match = re.search(r'\(([^)]+)\)', line)
                if val_match:
                    val_str = val_match.group(1)
                    vals = [float(x) for x in val_str.split()]
                else:
                    val_match = re.search(r'uniform\s+([0-9.eE+-]+)', line)
                    vals = [float(val_match.group(1))] if val_match else [0.0]
                
                if len(vals) == 1:
                    return np.full(shape, vals[0], dtype=np.float32)
                else:
                    arr = np.full((*shape, len(vals)), vals, dtype=np.float32)
                    return arr
            
            elif 'nonuniform' in line.lower():
                for j in range(i+1, min(i+5, len(lines))):
                    count_match = re.match(r'^\s*(\d+)\s*$', lines[j])
                    if count_match:
                        n_cells = int(count_match.group(1))
                        data_start = j + 1
                        break
            break
    
    if data_start is None:
        raise ValueError(f"Could not find data section in {field_path}")
    
    values = []
    paren_depth = 0
    current_vec = []
    
    for line in lines[data_start:]:
        line = line.strip()
        
        if line == ')' or line == ');':
            break
        if not line:
            continue
        
        if line.startswith('(') and ')' in line and line.index(')') == len(line) - 1:
            vec_str = line.strip('()')
            vals = [float(x) for x in vec_str.split() if x]
            if vals:
                values.append(vals)
        elif line.startswith('('):
            paren_depth = 1
            vec_str = line[1:]
            vals = [float(x) for x in vec_str.split() if x]
            current_vec.extend(vals)
        elif paren_depth > 0:
            if ')' in line:
                paren_depth = 0
                vec_str = line[:line.index(')')]
                vals = [float(x) for x in vec_str.split() if x]
                current_vec.extend(vals)
                if current_vec:
                    values.append(current_vec)
                current_vec = []
            else:
                vals = [float(x) for x in line.split() if x]
                current_vec.extend(vals)
        else:
            vals = [float(x) for x in line.split() if x]
            if vals:
                values.append(vals[0] if len(vals) == 1 else vals)
    
    if not values:
        raise ValueError(f"No data values found in {field_path}")
    
    arr = np.array(values, dtype=np.float32)
    expected_cells = shape[0] * shape[1]
    
    if arr.ndim == 1:
        if arr.size != expected_cells:
            print(f"    WARNING: Expected {expected_cells} cells, got {arr.size}. Truncating/padding.")
            if arr.size > expected_cells:
                arr = arr[:expected_cells]
            else:
                arr = np.pad(arr, (0, expected_cells - arr.size), mode='edge')
        return arr.reshape(shape).astype(np.float32)
    else:
        if arr.shape[0] != expected_cells:
            print(f"    WARNING: Expected {expected_cells} vectors, got {arr.shape[0]}. Truncating/padding.")
            if arr.shape[0] > expected_cells:
                arr = arr[:expected_cells]
            else:
                pad = np.zeros((expected_cells - arr.shape[0], arr.shape[1]), dtype=np.float32)
                arr = np.vstack([arr, pad])
        
        return arr.reshape((*shape, arr.shape[-1])).astype(np.float32)


def extract_case_spatial(case_dir, fields=['p', 'T', 'U', 'H2', 'O2'],
                          shape=(150, 300), max_timesteps=20):
    """Extract spatial fields for all time steps in a case."""
    case_path = Path(case_dir)
    
    time_dirs = []
    for d in sorted(case_path.iterdir()):
        if d.is_dir():
            try:
                t = float(d.name)
                if t > 1e-10:
                    time_dirs.append((t, d))
            except ValueError:
                continue
    
    time_dirs = sorted(time_dirs, key=lambda x: x[0])[:max_timesteps]
    
    if not time_dirs:
        raise ValueError(f"No valid time directories found in {case_dir}")
    
    snapshots = []
    times = []
    
    for t_val, t_dir in time_dirs:
        snapshot = {}
        valid = True
        
        for field in fields:
            field_file = t_dir / field
            
            try:
                arr = read_openfoam_field(field_file, shape)
                
                if field == 'U':
                    if arr.ndim == 3 and arr.shape[-1] >= 2:
                        snapshot['Ux'] = arr[..., 0]
                        snapshot['Uy'] = arr[..., 1]
                        if arr.shape[-1] > 2:
                            snapshot['Uz'] = arr[..., 2]
                    elif arr.ndim == 2:
                        snapshot['Ux'] = arr
                        snapshot['Uy'] = np.zeros_like(arr)
                    else:
                        print(f"    WARNING: Unexpected U shape {arr.shape} at t={t_val}")
                        valid = False
                        break
                else:
                    snapshot[field] = arr
                    
            except Exception as e:
                print(f"    ERROR reading {field} at t={t_val}: {e}")
                valid = False
                break
        
        if valid:
            channels = []
            channel_order = ['p', 'T', 'Ux', 'Uy', 'H2', 'O2']
            
            for key in channel_order:
                if key in snapshot:
                    ch = snapshot[key]
                    if ch.ndim != 2:
                        print(f"    WARNING: Channel {key} has shape {ch.shape}, expected 2D")
                        if ch.ndim == 3:
                            ch = ch[..., 0]
                    channels.append(ch)
                else:
                    print(f"    WARNING: Missing channel {key} at t={t_val}")
                    valid = False
                    break
            
            if valid and len(channels) == 6:
                shapes = [ch.shape for ch in channels]
                if len(set(shapes)) != 1:
                    print(f"    ERROR: Shape mismatch at t={t_val}: {dict(zip(channel_order, shapes))}")
                    valid = False
                else:
                    snapshot_arr = np.stack(channels, axis=0)
                    snapshots.append(snapshot_arr)
                    times.append(t_val)
        
        if not valid:
            print(f"    SKIPPED t={t_val:.6f}")
    
    if not snapshots:
        raise ValueError(f"No valid snapshots extracted from {case_dir}")
    
    return np.array(snapshots, dtype=np.float32), np.array(times, dtype=np.float32)


def load_case_params_from_csv(csv_path, case_name, col_mapping, df):
    """Load parameters using auto-detected column mapping."""
    if col_mapping is None or df is None:
        return None
    
    case_col = col_mapping.get('case_id')
    if case_col is None:
        return None
    
    # Try exact match first
    case_df = df[df[case_col].astype(str) == case_name]
    
    # If no match, try partial match
    if len(case_df) == 0:
        case_df = df[df[case_col].astype(str).str.contains(case_name, na=False)]
    
    if len(case_df) == 0:
        return None
    
    row = case_df.iloc[0]
    
    params = {'case_id': case_name}
    
    for std_name in ['phi', 'p0_pa', 'T0_k', 'cj_speed_ms']:
        csv_col = col_mapping.get(std_name)
        if csv_col and csv_col in row:
            params[std_name] = float(row[csv_col])
        else:
            # Fallback defaults
            defaults = {'phi': 1.0, 'p0_pa': 101325, 'T0_k': 300, 'cj_speed_ms': 1900}
            params[std_name] = defaults.get(std_name, 0.0)
    
    return params


def main():
    parser = argparse.ArgumentParser(description='Extract RDE spatial fields')
    parser.add_argument('--cases_dir', type=str, default='.',
                       help='Parent directory containing case folders')
    parser.add_argument('--cases', nargs='+', required=True,
                       help='Case folder names (e.g., RDE_01 RDE_02)')
    parser.add_argument('--csv', type=str, default='RDE_hybrid_dataset_v9.csv',
                       help='CSV file with case parameters')
    parser.add_argument('--output', type=str, default='spatial_fields.npz',
                       help='Output NPZ file')
    parser.add_argument('--shape', nargs=2, type=int, default=[150, 300],
                       help='Mesh dimensions H W')
    parser.add_argument('--max_t', type=int, default=20,
                       help='Max time steps per case')
    parser.add_argument('--fields', nargs='+',
                       default=['p', 'T', 'U', 'H2', 'O2'],
                       help='Fields to extract')
    args = parser.parse_args()
    
    all_fields = []
    all_times = []
    all_params = []
    
    print(f"Extracting fields: {args.fields}")
    print(f"Target shape: {tuple(args.shape)}")
    print(f"Max timesteps: {args.max_t}")
    print("=" * 70)
    
    # Load CSV and detect columns
    col_mapping = None
    df = None
    csv_path = Path(args.csv)
    
    if csv_path.exists():
        print(f"\nAnalyzing CSV: {args.csv}")
        try:
            col_mapping, df = sniff_csv_columns(csv_path)
            if 'case_id' not in col_mapping:
                print("  WARNING: Could not detect case ID column. Will use defaults.")
        except Exception as e:
            print(f"  ERROR reading CSV: {e}")
    else:
        print(f"\nWARNING: CSV not found: {args.csv}. Using default parameters.")
    
    for case in args.cases:
        case_dir = Path(args.cases_dir) / case
        print(f"\nProcessing {case}...")
        
        if not case_dir.exists():
            print(f"  ERROR: Directory not found: {case_dir}")
            continue
        
        try:
            fields, times = extract_case_spatial(
                case_dir,
                fields=args.fields,
                shape=tuple(args.shape),
                max_timesteps=args.max_t
            )
            
            # Load parameters
            params = load_case_params_from_csv(args.csv, case, col_mapping, df)
            if params is None:
                print(f"  WARNING: No params found for {case}, using defaults")
                params = {
                    'case_id': case,
                    'phi': 1.0,
                    'p0_pa': 101325,
                    'T0_k': 300,
                    'cj_speed_ms': 1900
                }
            
            print(f"  Extracted {len(times)} time steps, shape: {fields.shape}")
            print(f"    Time range: [{times.min():.6f}, {times.max():.6f}]")
            print(f"    Params: phi={params['phi']:.3f}, p0={params['p0_pa']:.0f}, "
                  f"T0={params['T0_k']:.0f}, Dcj={params['cj_speed_ms']:.1f}")
            
            all_fields.append(fields)
            all_times.append(times)
            all_params.append(params)
            
        except Exception as e:
            print(f"  ERROR processing {case}: {e}")
            import traceback
            traceback.print_exc()
            continue
    
    if not all_fields:
        print("\n" + "=" * 70)
        print("ERROR: No cases were successfully processed!")
        print("=" * 70)
        return
    
    all_fields = np.stack(all_fields, axis=0)
    all_times = np.stack(all_times, axis=0)
    
    print("\n" + "=" * 70)
    print(f"SAVED: {args.output}")
    print(f"  Shape: {all_fields.shape}")
    print(f"  Cases: {all_fields.shape[0]}")
    print(f"  Time steps: {all_fields.shape[1]}")
    print(f"  Channels: {all_fields.shape[2]}")
    print(f"  Spatial: {all_fields.shape[3]} x {all_fields.shape[4]}")
    print(f"  Size: {all_fields.nbytes / 1024**2:.1f} MB")
    print("=" * 70)
    
    np.savez_compressed(args.output,
                       fields=all_fields,
                       times=all_times,
                       params=json.dumps(all_params))


if __name__ == '__main__':
    main()