File size: 10,528 Bytes
e94bdab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
"""Generate conductivity-stratified benchmark splits for SSE screening tasks.

Extends the existing frozen splits with conductivity-aware splits:
  1. Conductivity held-out: hold out top 10% conductors by BVSE barrier
  2. Composition held-out + conducitivity-aware: no formula overlap 
  3. Family-stratified: balanced by SSE family

Usage:
    python scripts/generate_conductivity_splits.py                          # from existing splits + BVSE
    python scripts/generate_conductivity_splits.py --min-barrier 0.0        # include all
    python scripts/generate_conductivity_splits.py --dry-run                # stats only
"""
import json, os, sys, time, argparse, random
from pathlib import Path
from collections import defaultdict
import numpy as np

SEED = 42
SPLIT_DIR = "dataset/splits"


def main():
    parser = argparse.ArgumentParser(description="Generate conductivity-stratified splits")
    parser.add_argument("--min-barrier", type=float, default=0.0,
                        help="Minimum BVSE barrier to filter by (default: 0.0 = all)")
    parser.add_argument("--dry-run", action="store_true")
    parser.add_argument("--train-ratio", type=float, default=0.8)
    parser.add_argument("--val-ratio", type=float, default=0.1)
    parser.add_argument("--test-ratio", type=float, default=0.1)
    args = parser.parse_args()
    
    BASE_DIR = Path(__file__).resolve().parent.parent
    DATASET_PATH = BASE_DIR / "dataset"
    
    print("=" * 60)
    print("  CONDUCTIVITY BENCHMARK SPLITS")
    print("  Extending splits for SSE screening tasks")
    print("=" * 60)
    
    print("\nLoading entries...")
    t0 = time.time()
    with open(DATASET_PATH / "entries_final_v3.json") as f:
        entries = json.load(f)
    print(f"  {len(entries):,} entries ({time.time()-t0:.1f}s)")
    
    # Filter to entries with BVSE barrier data
    ssb_entries = [e for e in entries if e.get("ssb_screening", {}).get("bvse_migration_barrier_eV") is not None]
    print(f"\n  Entries with BVSE barriers: {len(ssb_entries):,}")
    
    if args.min_barrier > 0:
        ssb_entries = [e for e in ssb_entries 
                      if e["ssb_screening"]["bvse_migration_barrier_eV"] >= args.min_barrier]
        print(f"  After min-barrier {args.min_barrier:.1f} eV: {len(ssb_entries):,}")
    
    if not ssb_entries:
        print("  No entries with BVSE barriers found. Run compute_bvse_barriers.py first.")
        return
    
    random.seed(SEED)
    np.random.seed(SEED)
    
    splits = {}
    
    # 1. Conductivity-stratified split (stratified by BVSE barrier percentile)
    print(f"\n{'─' * 60}")
    print("  1. Conductivity-stratified split")
    print(f"{'─' * 60}")
    
    barriers = np.array([e["ssb_screening"]["bvse_migration_barrier_eV"] for e in ssb_entries])
    percentiles = np.percentile(barriers, [33, 67])
    
    low = [e for e in ssb_entries if e["ssb_screening"]["bvse_migration_barrier_eV"] <= percentiles[0]]
    mid = [e for e in ssb_entries if percentiles[0] < e["ssb_screening"]["bvse_migration_barrier_eV"] <= percentiles[1]]
    high = [e for e in ssb_entries if e["ssb_screening"]["bvse_migration_barrier_eV"] > percentiles[1]]
    
    print(f"    Low barrier (≤{percentiles[0]:.3f} eV): {len(low):,}")
    print(f"    Mid barrier ({percentiles[0]:.3f}-{percentiles[1]:.3f} eV): {len(mid):,}")
    print(f"    High barrier (≥{percentiles[1]:.3f} eV): {len(high):,}")
    
    stratified_train, stratified_val, stratified_test = [], [], []
    for pool in [low, mid, high]:
        np.random.shuffle(pool)
        n = len(pool)
        n_train = int(n * args.train_ratio)
        n_val = int(n * args.val_ratio)
        stratified_train.extend(pool[:n_train])
        stratified_val.extend(pool[n_train:n_train+n_val])
        stratified_test.extend(pool[n_train+n_val:])
    
    sorted_train = sorted(stratified_train, key=lambda e: e["ssb_screening"]["bvse_migration_barrier_eV"])
    sorted_val = sorted(stratified_val, key=lambda e: e["ssb_screening"]["bvse_migration_barrier_eV"])
    sorted_test = sorted(stratified_test, key=lambda e: e["ssb_screening"]["bvse_migration_barrier_eV"])
    
    splits["conductivity_stratified"] = {
        "train": [e["source_id"] + e.get("source", "") for e in sorted_train],
        "val": [e["source_id"] + e.get("source", "") for e in sorted_val],
        "test": [e["source_id"] + e.get("source", "") for e in sorted_test],
    }
    print(f"    Train: {len(splits['conductivity_stratified']['train']):,}")
    print(f"    Val: {len(splits['conductivity_stratified']['val']):,}")
    print(f"    Test: {len(splits['conductivity_stratified']['test']):,}")
    
    # 2. Family-stratified split (balanced by SSE family)
    print(f"\n{'─' * 60}")
    print("  2. Family-stratified split")
    print(f"{'─' * 60}")
    
    families = defaultdict(list)
    for e in ssb_entries:
        fam = e.get("ssb_screening", {}).get("sse_family") or e.get("sse_family", "unknown")
        families[fam].append(e)
    
    family_counts = {fam: len(entries) for fam, entries in sorted(families.items(), key=lambda x: -len(x[1]))}
    print(f"    Families: {len(families)}")
    for fam, count in list(family_counts.items())[:10]:
        print(f"      {fam:20s}: {count:,}")
    
    family_train, family_val, family_test = [], [], []
    for fam, pool in families.items():
        np.random.shuffle(pool)
        n = len(pool)
        n_train = max(1, int(n * args.train_ratio))
        n_val = max(1, int(n * args.val_ratio))
        family_train.extend(pool[:n_train])
        family_val.extend(pool[n_train:n_train+n_val])
        family_test.extend(pool[n_train+n_val:])
    
    splits["family_stratified_ssb"] = {
        "train": [e["source_id"] + e.get("source", "") for e in family_train],
        "val": [e["source_id"] + e.get("source", "") for e in family_val],
        "test": [e["source_id"] + e.get("source", "") for e in family_test],
    }
    print(f"    Train: {len(splits['family_stratified_ssb']['train']):,}")
    print(f"    Val: {len(splits['family_stratified_ssb']['val']):,}")
    print(f"    Test: {len(splits['family_stratified_ssb']['test']):,}")
    
    # 3. Best-candidate held-out (hold out top-100 lowest-barrier entries for testing)
    print(f"\n{'─' * 60}")
    print("  3. Best-candidate held-out split")
    print(f"{'─' * 60}")
    
    sorted_by_barrier = sorted(ssb_entries, key=lambda e: e["ssb_screening"]["bvse_migration_barrier_eV"])
    top_k = min(500, len(sorted_by_barrier))
    best_test = sorted_by_barrier[:top_k]
    best_pool = sorted_by_barrier[top_k:]
    
    np.random.shuffle(best_pool)
    n_best_train = int(len(best_pool) * args.train_ratio)
    n_best_val = int(len(best_pool) * args.val_ratio)
    best_train = best_pool[:n_best_train]
    best_val = best_pool[n_best_train:n_best_train+n_best_val]
    
    splits["best_conductors_held_out"] = {
        "train": [e["source_id"] + e.get("source", "") for e in best_train],
        "val": [e["source_id"] + e.get("source", "") for e in best_val],
        "test": [e["source_id"] + e.get("source", "") for e in best_test],
    }
    print(f"    Test (top {top_k} conductors): {len(splits['best_conductors_held_out']['test']):,}")
    print(f"    Train: {len(splits['best_conductors_held_out']['train']):,}")
    print(f"    Val: {len(splits['best_conductors_held_out']['val']):,}")
    
    # 4. Mobility class held-out (hold out entire mobility classes)
    print(f"\n{'─' * 60}")
    print("  4. Mobility-class held-out split")
    print(f"{'─' * 60}")
    
    classes = defaultdict(list)
    for e in ssb_entries:
        cls = e["ssb_screening"].get("bvse_mobility_class", "unknown")
        classes[cls].append(e)
    
    for cls, pool in classes.items():
        print(f"    {cls:15s}: {len(pool):,}")
    
    # Hold out superionic as test set (hardest generalization task)
    if "superionic" in classes:
        mob_test = classes["superionic"]
        mob_pool = []
        for cls, pool in classes.items():
            if cls != "superionic":
                mob_pool.extend(pool)
        np.random.shuffle(mob_pool)
        n_mob_train = int(len(mob_pool) * args.train_ratio)
        n_mob_val = int(len(mob_pool) * args.val_ratio)
        mob_train = mob_pool[:n_mob_train]
        mob_val = mob_pool[n_mob_train:n_mob_train+n_mob_val]
        
        splits["mobility_class_held_out"] = {
            "train": [e["source_id"] + e.get("source", "") for e in mob_train],
            "val": [e["source_id"] + e.get("source", "") for e in mob_val],
            "test": [e["source_id"] + e.get("source", "") for e in mob_test],
        }
        print(f"\n    Hold-out class: superionic ({len(mob_test):,} entries)")
        print(f"    Train: {len(mob_train):,}, Val: {len(mob_val):,}")
    else:
        print("    No superionic entries found for held-out split.")
    
    # Save splits
    if not args.dry_run:
        output_dir = BASE_DIR / SPLIT_DIR / "ssb"
        output_dir.mkdir(parents=True, exist_ok=True)
        
        for split_name, split_data in splits.items():
            output_path = output_dir / f"{split_name}.json"
            
            # Convert to index-based splits
            entry_indices = {}
            for i, e in enumerate(entries):
                key = e["source_id"] + e.get("source", "")
                entry_indices[key] = i
            
            index_split = {
                "train": [entry_indices[k] for k in split_data["train"] if k in entry_indices],
                "val": [entry_indices[k] for k in split_data["val"] if k in entry_indices],
                "test": [entry_indices[k] for k in split_data["test"] if k in entry_indices],
            }
            
            with open(output_path, "w") as f:
                json.dump(index_split, f, indent=2)
            print(f"\n  Saved: {output_path}")
            print(f"    Train: {len(index_split['train']):,}")
            print(f"    Val: {len(index_split['val']):,}")
            print(f"    Test: {len(index_split['test']):,}")
        
        # Summary
        print(f"\n{'─' * 60}")
        print("  SPLIT SUMMARY")
        print(f"{'─' * 60}")
        for split_name in splits:
            data = splits[split_name]
            print(f"  {split_name:35s}: train={len(data['train']):,} val={len(data['val']):,} test={len(data['test']):,}")
    
    else:
        print(f"\n  (dry-run — no files written)")
    
    print("=" * 60)


if __name__ == "__main__":
    main()