Spaces:
Running
Running
File size: 7,437 Bytes
d29b763 | 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 | """Ablation study for the enriched MEDCARE-DDI feature pipeline.
Runs four arms:
- DDInter only
- DDInter + DrugBank
- DDInter + TWOSIDES
- Full pipeline
Outputs:
- CSV summary
- Markdown report
- bar chart
- confusion matrix PNG per arm
- JSON summary
"""
from __future__ import annotations
import json
import sys
import gc
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
ROOT = Path(__file__).resolve().parents[2]
sys.path.insert(0, str(ROOT / 'src'))
from training.feature_pipeline import build_feature_pipeline
from training.retrain_production_model import TrainConfig, train_and_evaluate
REPORT_DIR = ROOT / 'models' / 'reports'
REPORT_DIR.mkdir(parents=True, exist_ok=True)
def _stratified_sample(df: pd.DataFrame, sample_size: int, seed: int = 2026) -> pd.DataFrame:
if sample_size <= 0 or sample_size >= len(df):
return df.copy()
group_cols = ['label']
grouped = df.groupby(group_cols, group_keys=False)
fractions = min(1.0, sample_size / float(len(df)))
sampled = grouped.apply(lambda part: part.sample(frac=fractions, random_state=seed))
if len(sampled) > sample_size:
sampled = sampled.sample(n=sample_size, random_state=seed)
return sampled.reset_index(drop=True)
def _mask_groups(X: np.ndarray, group_slices: dict[str, tuple[int, int]], enabled_groups: list[str]) -> np.ndarray:
masked = np.zeros_like(X)
for group_name in enabled_groups:
if group_name not in group_slices:
continue
start, end = group_slices[group_name]
masked[:, start:end] = X[:, start:end]
return masked
def _save_confusion_matrix(cm: list[list[int]], labels: list[str], out_path: Path) -> None:
matrix = np.asarray(cm, dtype=np.int64)
fig, ax = plt.subplots(figsize=(6, 5))
im = ax.imshow(matrix, cmap='Blues')
fig.colorbar(im, ax=ax)
ax.set_xticks(range(len(labels)))
ax.set_yticks(range(len(labels)))
ax.set_xticklabels(labels, rotation=45, ha='right')
ax.set_yticklabels(labels)
ax.set_xlabel('Predicted')
ax.set_ylabel('True')
ax.set_title('Confusion Matrix')
threshold = matrix.max() / 2.0 if matrix.size else 0
for i in range(matrix.shape[0]):
for j in range(matrix.shape[1]):
ax.text(j, i, str(matrix[i, j]), ha='center', va='center', color='white' if matrix[i, j] > threshold else 'black')
fig.tight_layout()
fig.savefig(out_path, dpi=160)
plt.close(fig)
def _markdown_table(df: pd.DataFrame) -> str:
headers = list(df.columns)
rows = [headers]
for _, row in df.iterrows():
rows.append([str(row[col]) for col in headers])
widths = [max(len(cell) for cell in column) for column in zip(*rows)]
lines = []
lines.append('| ' + ' | '.join(header.ljust(widths[idx]) for idx, header in enumerate(headers)) + ' |')
lines.append('| ' + ' | '.join('-' * widths[idx] for idx in range(len(headers))) + ' |')
for row in rows[1:]:
lines.append('| ' + ' | '.join(cell.ljust(widths[idx]) for idx, cell in enumerate(row)) + ' |')
return '\n'.join(lines)
def main() -> None:
print('Building shared feature pipeline...', flush=True)
pairs_df, artifacts = build_feature_pipeline(save_artifacts=True, sample_size=3000, seed=2026)
group_slices = artifacts.group_slices
X = np.asarray(list(pairs_df['_X'].values), dtype=np.float32)
base_groups = ['pair_encoding', 'semantic_embeddings', 'normalized_numeric']
arms = {
'ddinter_only': base_groups,
'ddinter_drugbank': base_groups + ['drugbank_active', 'drugbank_atc', 'drugbank_category'],
'ddinter_twosides': base_groups + ['twosides_signal'],
'full': base_groups + ['drugbank_active', 'drugbank_atc', 'drugbank_category', 'twosides_signal', 'polypharmacy'],
}
results: list[dict[str, object]] = []
summary_by_arm: dict[str, dict[str, object]] = {}
for arm_name, enabled_groups in arms.items():
arm_X = _mask_groups(X, group_slices, enabled_groups)
arm_df = pairs_df[['drug_a', 'drug_b', 'label', 'support']].copy()
arm_df['_X'] = list(arm_X.tolist())
train_df, temp_df = train_test_split(arm_df, test_size=0.2, stratify=arm_df['label'], random_state=2026)
val_df, test_df = train_test_split(temp_df, test_size=0.5, stratify=temp_df['label'], random_state=2026)
config = TrainConfig(
seed=2026,
embedding_dim=32,
hidden_dim=64,
dropout=0.15,
lr=1e-3,
batch_size=128,
weight_decay=1e-5,
epochs=2,
loss_type='focal',
sampler='weighted',
class_weights=[],
)
print(f'Running arm={arm_name} with groups={enabled_groups} and samples={len(arm_df)}', flush=True)
report = train_and_evaluate(config, train_df, val_df, test_df, vocab={})
summary = {
'arm': arm_name,
'accuracy': report['accuracy'],
'macro_f1': report['macro_f1'],
'severe_recall': report['severe_recall'],
'num_test_examples': report['num_test_examples'],
'enabled_groups': enabled_groups,
}
results.append(summary)
summary_by_arm[arm_name] = report
cm_path = REPORT_DIR / f'ablation_confusion_matrix_{arm_name}.png'
_save_confusion_matrix(report['confusion_matrix'], report['label_names'], cm_path)
del arm_X, arm_df, train_df, val_df, test_df, report
gc.collect()
summary_df = pd.DataFrame(results).sort_values(by=['severe_recall', 'macro_f1'], ascending=False)
summary_csv = REPORT_DIR / 'ablation_summary.csv'
summary_df.to_csv(summary_csv, index=False)
summary_json = REPORT_DIR / 'ablation_summary.json'
summary_json.write_text(json.dumps(results, indent=2), encoding='utf-8')
chart_path = REPORT_DIR / 'ablation_metrics.png'
fig, axes = plt.subplots(1, 3, figsize=(14, 4), sharex=True)
for ax, metric in zip(axes, ['accuracy', 'macro_f1', 'severe_recall']):
ax.bar(summary_df['arm'], summary_df[metric], color=['#4C78A8', '#72B7B2', '#F58518', '#54A24B'])
ax.set_title(metric.replace('_', ' ').title())
ax.set_ylim(0, 1)
ax.tick_params(axis='x', rotation=20)
fig.tight_layout()
fig.savefig(chart_path, dpi=160)
plt.close(fig)
report_md = REPORT_DIR / 'ablation_report.md'
lines = [
'# Ablation Study Report',
'',
'## Summary',
'',
_markdown_table(summary_df),
'',
'## Interpretation',
'',
'- DDInter-only is the baseline arm.',
'- DrugBank arm adds active ingredients, ATC codes, and categories.',
'- TWOSIDES arm adds side-effect signal overlap.',
'- Full pipeline includes all feature groups and should perform best if the feature integration is correct.',
'',
'## Artifacts',
'',
f'- CSV: {summary_csv}',
f'- JSON: {summary_json}',
f'- Chart: {chart_path}',
]
report_md.write_text('\n'.join(lines), encoding='utf-8')
print('Ablation complete.')
print(f'Summary CSV: {summary_csv}')
print(f'Summary JSON: {summary_json}')
print(f'Report: {report_md}')
print(f'Chart: {chart_path}')
if __name__ == '__main__':
main()
|