oncall-guide-ai / dataset /scripts /data_explorer_treatment.py
YanBoChen
WIP: add dual keyword and text length distribution plots for treatment subset analysis
a5bcfa7
raw
history blame
13.8 kB
# /scripts/data_explorer_treatment.py
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from pathlib import Path
import json
import numpy as np
from tqdm import tqdm
import re
def analyze_treatment_subset(
treatment_file_path,
emergency_keywords_path,
treatment_keywords_path,
output_dir="analysis_treatment"
):
"""
Specialized analysis for treatment subset focusing on:
1. Dual keyword analysis (emergency + treatment)
2. Path B effectiveness validation
3. Condition mapping data preparation
4. RAG readiness assessment
"""
print(f"\n{'='*60}")
print(f"Treatment Subset Analysis")
print(f"Treatment file: {treatment_file_path}")
print(f"Emergency keywords: {emergency_keywords_path}")
print(f"Treatment keywords: {treatment_keywords_path}")
print(f"Output directory: {output_dir}")
print(f"{'='*60}\n")
# Load data
print("1️⃣ Loading treatment subset data...")
df = pd.read_csv(treatment_file_path)
output_dir = Path(output_dir)
# Load keyword lists
print("2️⃣ Loading keyword lists...")
with open(emergency_keywords_path, 'r', encoding='utf-8') as f:
emergency_keywords = [line.strip() for line in f if line.strip()]
with open(treatment_keywords_path, 'r', encoding='utf-8') as f:
treatment_keywords = [line.strip() for line in f if line.strip()]
print(f" Emergency keywords: {len(emergency_keywords)}")
print(f" Treatment keywords: {len(treatment_keywords)}")
# Basic statistics
print("\n3️⃣ Computing basic statistics...")
total_records = len(df)
df['text_length'] = df['clean_text'].str.len()
avg_length = df['text_length'].mean()
print(f" Total treatment records: {total_records}")
print(f" Average text length: {avg_length:.2f} characters")
# Initialize comprehensive statistics
stats = {
'basic_statistics': {
'total_records': int(total_records),
'avg_text_length': float(avg_length),
'emergency_keywords_count': len(emergency_keywords),
'treatment_keywords_count': len(treatment_keywords)
},
'emergency_keyword_stats': {},
'treatment_keyword_stats': {},
'cooccurrence_analysis': {},
'path_b_validation': {},
'condition_mapping_candidates': {}
}
# Emergency keyword analysis in treatment subset
print("\n4️⃣ Analyzing emergency keywords in treatment subset...")
for keyword in emergency_keywords:
count = df['clean_text'].str.contains(keyword, case=False, na=False).sum()
stats['emergency_keyword_stats'][keyword] = int(count)
print(f" Emergency: {keyword} -> {count} records")
# Treatment keyword analysis
print("\n5️⃣ Analyzing treatment keywords...")
for keyword in treatment_keywords:
count = df['clean_text'].str.contains(keyword, case=False, na=False).sum()
stats['treatment_keyword_stats'][keyword] = int(count)
print(f" Treatment: {keyword} -> {count} records")
# Step 6: Co-occurrence analysis
print("\n6️⃣ Computing keyword co-occurrence patterns...")
# Initialize matrices for full dataset
emergency_matrix = np.zeros((len(df), len(emergency_keywords)), dtype=bool)
treatment_matrix = np.zeros((len(df), len(treatment_keywords)), dtype=bool)
# Pre-process text
print(" Pre-processing text...")
df['clean_text_lower'] = df['clean_text'].fillna('').str.lower()
# Process all emergency keywords
print("\n Processing all emergency keywords...")
for i, keyword in enumerate(tqdm(emergency_keywords, desc="Emergency keywords")):
pattern = r'(?<!\w)' + re.escape(keyword) + r'(?!\w)'
emergency_matrix[:, i] = df['clean_text_lower'].str.contains(pattern, regex=True, na=False)
matches = emergency_matrix[:, i].sum()
print(f" - {keyword}: {matches} matches")
# Process all treatment keywords
print("\n Processing all treatment keywords...")
for i, keyword in enumerate(tqdm(treatment_keywords, desc="Treatment keywords")):
pattern = r'(?<!\w)' + re.escape(keyword) + r'(?!\w)'
treatment_matrix[:, i] = df['clean_text_lower'].str.contains(pattern, regex=True, na=False)
matches = treatment_matrix[:, i].sum()
print(f" - {keyword}: {matches} matches")
# Compute co-occurrence matrix
print("\n Computing co-occurrence matrix...")
cooc_matrix = emergency_matrix.astype(int).T @ treatment_matrix.astype(int)
print(" Computation completed successfully")
# Extract results
print(" Extracting co-occurrence pairs...")
cooccurrence_pairs = []
for i, em_kw in enumerate(emergency_keywords):
for j, tr_kw in enumerate(treatment_keywords):
count = int(cooc_matrix[i, j])
if count > 0:
cooccurrence_pairs.append({
'emergency_keyword': em_kw,
'treatment_keyword': tr_kw,
'cooccurrence_count': count,
'percentage': float(count / len(df) * 100)
})
# Sort and store results
cooccurrence_pairs.sort(key=lambda x: x['cooccurrence_count'], reverse=True)
stats['cooccurrence_analysis'] = cooccurrence_pairs[:20] # Top 20 pairs
print(f" Found {len(cooccurrence_pairs)} co-occurrence pairs")
print(" Top 5 co-occurrence pairs:")
for i, pair in enumerate(cooccurrence_pairs[:5]):
print(f" {i+1}. {pair['emergency_keyword']} + {pair['treatment_keyword']}: {pair['cooccurrence_count']} ({pair['percentage']:.1f}%)")
# Step 7: Path B validation metrics
print("\n7️⃣ Validating Path B strategy effectiveness...")
# Compute keyword density with progress bar
print(" Computing keyword density...")
with tqdm(total=2, desc="Density calculation") as pbar:
emergency_density = emergency_matrix.sum(axis=1)
pbar.update(1)
treatment_density = treatment_matrix.sum(axis=1)
pbar.update(1)
# Store density in dataframe
df['emergency_keyword_density'] = emergency_density
df['treatment_keyword_density'] = treatment_density
# Calculate statistics
stats['path_b_validation'] = {
'avg_emergency_density': float(np.mean(emergency_density)),
'avg_treatment_density': float(np.mean(treatment_density)),
'high_density_records': int(sum((emergency_density >= 2) & (treatment_density >= 2))),
'precision_estimate': float(sum((emergency_density >= 1) & (treatment_density >= 1)) / len(df))
}
# Print detailed results
print("\n Results:")
print(f" - Average emergency keyword density: {stats['path_b_validation']['avg_emergency_density']:.2f}")
print(f" - Average treatment keyword density: {stats['path_b_validation']['avg_treatment_density']:.2f}")
print(f" - High-density records (≥2 each): {stats['path_b_validation']['high_density_records']}")
print(f" - Precision estimate: {stats['path_b_validation']['precision_estimate']:.2f}")
# Sample distribution analysis
print("\n Density Distribution:")
density_counts = pd.DataFrame({
'emergency': emergency_density,
'treatment': treatment_density
}).value_counts().head()
print(" Top 5 density combinations (emergency, treatment):")
for (em, tr), count in density_counts.items():
print(f" - {count} documents have {em} emergency and {tr} treatment keywords")
# Condition mapping candidates
print("\n8️⃣ Preparing condition mapping candidates...")
# Group emergency keywords by potential conditions
condition_candidates = {}
for pair in cooccurrence_pairs[:10]: # Top 10 pairs
em_kw = pair['emergency_keyword']
tr_kw = pair['treatment_keyword']
# Simple condition inference (can be enhanced later)
if any(cardiac_term in em_kw.lower() for cardiac_term in ['mi', 'cardiac', 'heart', 'chest']):
condition = 'cardiac'
elif any(resp_term in em_kw.lower() for resp_term in ['respiratory', 'breathing', 'lung', 'dyspnea']):
condition = 'respiratory'
elif any(neuro_term in em_kw.lower() for neuro_term in ['stroke', 'seizure', 'consciousness']):
condition = 'neurological'
else:
condition = 'general'
if condition not in condition_candidates:
condition_candidates[condition] = []
condition_candidates[condition].append({
'emergency_keyword': em_kw,
'treatment_keyword': tr_kw,
'strength': pair['cooccurrence_count']
})
stats['condition_mapping_candidates'] = condition_candidates
# Visualization
print("\n9️⃣ Generating visualizations...")
output_plots = output_dir / "plots"
output_plots.mkdir(parents=True, exist_ok=True)
# 1. Dual keyword distribution
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(20, 8))
# Emergency keywords in treatment subset
em_counts = list(stats['emergency_keyword_stats'].values())
em_labels = list(stats['emergency_keyword_stats'].keys())
ax1.bar(range(len(em_labels)), em_counts)
ax1.set_title('Emergency Keywords in Treatment Subset')
ax1.set_xlabel('Emergency Keywords')
ax1.set_ylabel('Document Count')
ax1.tick_params(axis='x', rotation=45, labelsize=8)
ax1.set_xticks(range(len(em_labels)))
ax1.set_xticklabels(em_labels, ha='right')
# Treatment keywords
tr_counts = list(stats['treatment_keyword_stats'].values())
tr_labels = list(stats['treatment_keyword_stats'].keys())
ax2.bar(range(len(tr_labels)), tr_counts)
ax2.set_title('Treatment Keywords Distribution')
ax2.set_xlabel('Treatment Keywords')
ax2.set_ylabel('Document Count')
ax2.tick_params(axis='x', rotation=45, labelsize=8)
ax2.set_xticks(range(len(tr_labels)))
ax2.set_xticklabels(tr_labels, ha='right')
plt.tight_layout()
plt.savefig(output_plots / "dual_keyword_distribution.png", bbox_inches='tight', dpi=300)
plt.close()
# 2. Co-occurrence heatmap (top pairs)
if len(cooccurrence_pairs) > 0:
top_pairs = cooccurrence_pairs[:15] # Top 15 for readability
cooc_matrix = np.zeros((len(set([p['emergency_keyword'] for p in top_pairs])),
len(set([p['treatment_keyword'] for p in top_pairs]))))
em_unique = list(set([p['emergency_keyword'] for p in top_pairs]))
tr_unique = list(set([p['treatment_keyword'] for p in top_pairs]))
for pair in top_pairs:
i = em_unique.index(pair['emergency_keyword'])
j = tr_unique.index(pair['treatment_keyword'])
cooc_matrix[i, j] = pair['cooccurrence_count']
plt.figure(figsize=(12, 8))
sns.heatmap(cooc_matrix,
xticklabels=tr_unique,
yticklabels=em_unique,
annot=True,
fmt='g',
cmap='YlOrRd')
plt.title('Emergency-Treatment Keywords Co-occurrence Heatmap')
plt.xlabel('Treatment Keywords')
plt.ylabel('Emergency Keywords')
plt.xticks(rotation=45, ha='right')
plt.yticks(rotation=0)
plt.tight_layout()
plt.savefig(output_plots / "cooccurrence_heatmap.png", bbox_inches='tight', dpi=300)
plt.close()
# 3. Text length distribution
plt.figure(figsize=(10, 6))
df['text_length'].hist(bins=50, alpha=0.7)
plt.title('Text Length Distribution in Treatment Subset')
plt.xlabel('Text Length (characters)')
plt.ylabel('Frequency')
plt.axvline(avg_length, color='red', linestyle='--', label=f'Average: {avg_length:.0f}')
plt.legend()
plt.savefig(output_plots / "text_length_distribution.png", bbox_inches='tight')
plt.close()
# 4. Keyword density scatter plot
plt.figure(figsize=(10, 8))
plt.scatter(df['emergency_keyword_density'], df['treatment_keyword_density'], alpha=0.6)
plt.xlabel('Emergency Keyword Density')
plt.ylabel('Treatment Keyword Density')
plt.title('Emergency vs Treatment Keyword Density')
plt.grid(True, alpha=0.3)
plt.savefig(output_plots / "keyword_density_scatter.png", bbox_inches='tight')
plt.close()
# Save comprehensive statistics
print("\n🔟 Saving analysis results...")
stats_dir = output_dir / "stats"
stats_dir.mkdir(parents=True, exist_ok=True)
with open(stats_dir / "treatment_analysis_comprehensive.json", 'w', encoding='utf-8') as f:
json.dump(stats, f, indent=2, ensure_ascii=False)
# Save co-occurrence pairs as CSV for easy review
if cooccurrence_pairs:
cooc_df = pd.DataFrame(cooccurrence_pairs)
cooc_df.to_csv(stats_dir / "cooccurrence_pairs.csv", index=False)
print(f"✅ Treatment subset analysis complete!")
print(f" Results saved to: {output_dir}")
print(f" Plots: {output_plots}")
print(f" Statistics: {stats_dir}")
return stats
if __name__ == "__main__":
# Configuration
treatment_file = "../dataset/emergency_treatment/emergency_treatment_subset.csv"
emergency_keywords = "../keywords/emergency_keywords.txt"
treatment_keywords = "../keywords/treatment_keywords.txt"
output_directory = "../analysis_treatment"
# Run analysis
results = analyze_treatment_subset(
treatment_file,
emergency_keywords,
treatment_keywords,
output_directory
)