Spaces:
Running
Running
File size: 19,713 Bytes
c2ea5ed |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 |
#!/usr/bin/env python3
"""
DoWhy Causal Component Analysis
This script implements causal inference methods using the DoWhy library to analyze
the causal relationship between knowledge graph components and perturbation scores.
"""
import os
import sys
import pandas as pd
import numpy as np
import argparse
import logging
import json
from typing import Dict, List, Optional, Tuple, Set
from collections import defaultdict
# Import DoWhy
import dowhy
from dowhy import CausalModel
# Import from utils directory
from .utils.dataframe_builder import create_component_influence_dataframe
# Import shared utilities
from .utils.shared_utils import create_mock_perturbation_scores, list_available_components
# Configure logging
logger = logging.getLogger(__name__)
# Suppress DoWhy/info logs by setting their loggers to WARNING or higher
logging.basicConfig(level=logging.CRITICAL, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
# Suppress DoWhy and related noisy loggers
for noisy_logger in [
"dowhy",
"dowhy.causal_estimator",
"dowhy.causal_model",
"dowhy.causal_refuter",
"dowhy.do_sampler",
"dowhy.identifier",
"dowhy.propensity_score",
"dowhy.utils",
"dowhy.causal_refuter.add_unobserved_common_cause"
]:
logging.getLogger(noisy_logger).setLevel(logging.WARNING)
# Note: create_mock_perturbation_scores and list_available_components
# moved to utils.shared_utils to avoid duplication
def generate_simple_causal_graph(df: pd.DataFrame, treatment: str, outcome: str) -> str:
"""
Generate a simple causal graph in a format compatible with DoWhy.
Args:
df: DataFrame with features
treatment: Treatment variable name
outcome: Outcome variable name
Returns:
String representation of the causal graph in DoWhy format
"""
# Get component columns (all other variables that could affect both treatment and outcome)
component_cols = [col for col in df.columns if col.startswith(('entity_', 'relation_')) and col != treatment]
# Identify potential confounders by checking correlation patterns with the treatment
confounder_threshold = 0.7 # Correlation threshold to identify potential confounders
potential_confounders = []
# Calculate correlations between components to identify potential confounders
# A high correlation may indicate a confounder relationship
for component in component_cols:
# Skip if no variance (would result in correlation NaN)
if df[component].std() == 0 or df[treatment].std() == 0:
continue
correlation = df[component].corr(df[treatment])
if abs(correlation) >= confounder_threshold:
potential_confounders.append(component)
# Create a graph in DOT format
graph = "digraph {"
# Add edges for Treatment -> Outcome
graph += f'"{treatment}" -> "{outcome}";'
# Add edges for identified confounders
for confounder in potential_confounders:
# Confounder affects both treatment and outcome
graph += f'"{confounder}" -> "{treatment}";'
graph += f'"{confounder}" -> "{outcome}";'
# For remaining components (non-confounders), we'll add them as potential causes of the outcome
# but not necessarily related to the treatment
for component in component_cols:
if component not in potential_confounders:
graph += f'"{component}" -> "{outcome}";'
graph += "}"
return graph
def run_dowhy_analysis(
df: pd.DataFrame,
treatment_component: str,
outcome_var: str = "perturbation",
proceed_when_unidentifiable: bool = True
) -> Dict:
"""
Run causal analysis using DoWhy for a single treatment component.
Args:
df: DataFrame with binary component features and outcome variable
treatment_component: Name of the component to analyze
outcome_var: Name of the outcome variable
proceed_when_unidentifiable: Whether to proceed when effect is unidentifiable
Returns:
Dictionary with causal analysis results
"""
# Ensure the treatment_component is in the expected format
if treatment_component in df.columns:
treatment = treatment_component
else:
logger.error(f"Treatment component {treatment_component} not found in DataFrame")
return {"component": treatment_component, "error": f"Component not found"}
# Check for potential interaction effects with other components
interaction_components = []
# Look for potential interaction effects
# An interaction effect might be present if two variables together have a different effect
# than the sum of their individual effects
if df[treatment].sum() > 0: # Only check if the treatment appears in the data
# Get other components to check for interactions
other_components = [col for col in df.columns if col.startswith(('entity_', 'relation_'))
and col != treatment and col != outcome_var]
for component in other_components:
# Skip components with no occurrences
if df[component].sum() == 0:
continue
# Check if the component co-occurs with the treatment more than expected by chance
# This is a simplistic approach to identify potential interactions
expected_cooccurrence = (df[treatment].mean() * df[component].mean()) * len(df)
actual_cooccurrence = (df[treatment] & df[component]).sum()
# If actual co-occurrence is significantly different from expected
if actual_cooccurrence > 1.5 * expected_cooccurrence:
interaction_components.append(component)
# Generate a simple causal graph
graph = generate_simple_causal_graph(df, treatment, outcome_var)
# Create the causal model
try:
model = CausalModel(
data=df,
treatment=treatment,
outcome=outcome_var,
graph=graph,
proceed_when_unidentifiable=proceed_when_unidentifiable
)
# Print the graph (for debugging)
logger.info(f"Causal graph for {treatment}: {graph}")
# Identify the causal effect
identified_estimand = model.identify_effect(proceed_when_unidentifiable=proceed_when_unidentifiable)
logger.info(f"Identified estimand for {treatment}")
# If there's no variance in the outcome, we can't estimate effect
if df[outcome_var].std() == 0:
logger.warning(f"No variance in outcome variable {outcome_var}, skipping estimation")
return {
"component": treatment.replace("comp_", ""),
"identified_estimand": str(identified_estimand),
"error": "No variance in outcome variable"
}
# Estimate the causal effect
try:
estimate = model.estimate_effect(
identified_estimand,
method_name="backdoor.linear_regression",
target_units="ate",
test_significance=None
)
logger.info(f"Estimated causal effect for {treatment}: {estimate.value}")
# Check for interaction effects if we found potential interaction components
interaction_effects = []
if interaction_components:
for interaction_component in interaction_components:
# Create interaction term (product of both components)
interaction_col = f"{treatment}_x_{interaction_component}"
df[interaction_col] = df[treatment] * df[interaction_component]
# Run a simple linear regression with the interaction term
X = df[[treatment, interaction_component, interaction_col]]
y = df[outcome_var]
try:
from sklearn.linear_model import LinearRegression
model_with_interaction = LinearRegression()
model_with_interaction.fit(X, y)
# Get the coefficient for the interaction term
interaction_coef = model_with_interaction.coef_[2] # Index 2 is the interaction term
# Store the interaction effect
interaction_effects.append({
"component": interaction_component,
"interaction_coefficient": float(interaction_coef)
})
# Clean up temporary column
df.drop(columns=[interaction_col], inplace=True)
except Exception as e:
logger.warning(f"Error analyzing interaction with {interaction_component}: {str(e)}")
# Refute the results
refutation_results = []
# 1. Random common cause refutation
try:
rcc_refute = model.refute_estimate(
identified_estimand,
estimate,
method_name="random_common_cause"
)
refutation_results.append({
"method": "random_common_cause",
"refutation_result": str(rcc_refute)
})
except Exception as e:
logger.warning(f"Random common cause refutation failed: {str(e)}")
# 2. Placebo treatment refutation
try:
placebo_refute = model.refute_estimate(
identified_estimand,
estimate,
method_name="placebo_treatment_refuter"
)
refutation_results.append({
"method": "placebo_treatment",
"refutation_result": str(placebo_refute)
})
except Exception as e:
logger.warning(f"Placebo treatment refutation failed: {str(e)}")
# 3. Data subset refutation
try:
subset_refute = model.refute_estimate(
identified_estimand,
estimate,
method_name="data_subset_refuter"
)
refutation_results.append({
"method": "data_subset",
"refutation_result": str(subset_refute)
})
except Exception as e:
logger.warning(f"Data subset refutation failed: {str(e)}")
result = {
"component": treatment,
"identified_estimand": str(identified_estimand),
"effect_estimate": float(estimate.value),
"refutation_results": refutation_results
}
# Add interaction effects if found
if interaction_effects:
result["interaction_effects"] = interaction_effects
return result
except Exception as e:
logger.error(f"Error estimating effect for {treatment}: {str(e)}")
return {
"component": treatment,
"identified_estimand": str(identified_estimand),
"error": f"Estimation error: {str(e)}"
}
except Exception as e:
logger.error(f"Error in causal analysis for {treatment}: {str(e)}")
return {
"component": treatment,
"error": str(e)
}
def analyze_components_with_dowhy(
df: pd.DataFrame,
components_to_analyze: List[str]
) -> List[Dict]:
"""
Analyze causal effects of multiple components using DoWhy.
Args:
df: DataFrame with binary component features and outcome variable
components_to_analyze: List of component names to analyze
Returns:
List of dictionaries with causal analysis results
"""
results = []
# Track relationships between components for post-processing
interaction_map = defaultdict(list)
confounder_map = defaultdict(list)
# First, analyze each component individually
for component in components_to_analyze:
print(f"\nAnalyzing causal effect of component: {component}")
result = run_dowhy_analysis(df, component)
results.append(result)
# Print result summary
if "error" in result:
print(f" Error: {result['error']}")
else:
print(f" Estimated causal effect: {result.get('effect_estimate', 'N/A')}")
# Track interactions if found
if "interaction_effects" in result:
for interaction in result["interaction_effects"]:
interacting_component = interaction["component"]
interaction_coef = interaction["interaction_coefficient"]
# Record the interaction effect
interaction_entry = {
"component": component,
"interaction_coefficient": interaction_coef
}
interaction_map[interacting_component].append(interaction_entry)
print(f" Interaction with {interacting_component}: {interaction_coef}")
# Post-process to identify components that consistently appear in interactions
# or as confounders
for result in results:
component = result.get("component")
# Skip results with errors
if "error" in result or not component:
continue
# Add interactions information to the result
if component in interaction_map and interaction_map[component]:
result["interacts_with"] = interaction_map[component]
return results
def main():
"""Main function to run the DoWhy causal component analysis."""
# Set up argument parser
parser = argparse.ArgumentParser(description='DoWhy Causal Component Analysis')
parser.add_argument('--test', action='store_true', help='Enable test mode with mock perturbation scores')
parser.add_argument('--components', nargs='+', help='Component names to test in test mode')
parser.add_argument('--treatments', nargs='+', help='Component names to treat as treatments for causal analysis')
parser.add_argument('--list-components', action='store_true', help='List available components and exit')
parser.add_argument('--base-score', type=float, default=1.0, help='Base perturbation score (default: 1.0)')
parser.add_argument('--treatment-score', type=float, default=0.2, help='Score for test components (default: 0.2)')
parser.add_argument('--json-file', type=str, help='Path to JSON file (default: example.json)')
parser.add_argument('--top-k', type=int, default=5, help='Number of top components to analyze (default: 5)')
args = parser.parse_args()
# Path to example.json file or user-specified file
if args.json_file:
json_file = args.json_file
else:
json_file = os.path.join(os.path.dirname(__file__), 'example.json')
# Create DataFrame using the function from create_component_influence_dataframe.py
df = create_component_influence_dataframe(json_file)
if df is None or df.empty:
logger.error("Failed to create or empty DataFrame. Cannot proceed with analysis.")
return
# List components if requested
if args.list_components:
components = list_available_components(df)
print("\nAvailable components:")
for i, comp in enumerate(components, 1):
print(f"{i}. {comp}")
return
# Create mock perturbation scores if in test mode
if args.test:
if not args.components:
logger.warning("No components specified for test mode. Using random components.")
# Select random components if none specified
all_components = list_available_components(df)
if len(all_components) > 0:
test_components = np.random.choice(all_components,
size=min(2, len(all_components)),
replace=False).tolist()
else:
logger.error("No components found in DataFrame. Cannot create mock scores.")
return
else:
test_components = args.components
print(f"\nTest mode enabled. Using components: {', '.join(test_components)}")
print(f"Setting base score: {args.base_score}, treatment score: {args.treatment_score}")
# Create mock perturbation scores
df = create_mock_perturbation_scores(
df,
test_components,
base_score=args.base_score,
treatment_score=args.treatment_score
)
# Print basic DataFrame info
print(f"\nDataFrame info:")
print(f"Rows: {len(df)}")
feature_cols = [col for col in df.columns if col.startswith("comp_")]
print(f"Features: {len(feature_cols)}")
print(f"Columns: {', '.join([col for col in df.columns if not col.startswith('comp_')])}")
# Check if we have any variance in perturbation scores
if df['perturbation'].std() == 0:
print("\nWARNING: All perturbation scores are identical (value: %.2f)." % df['perturbation'].iloc[0])
print(" This will limit the effectiveness of causal analysis.")
print(" Consider using synthetic data with varied perturbation scores for better results.\n")
else:
print(f"\nPerturbation score statistics:")
print(f"Min: {df['perturbation'].min():.2f}")
print(f"Max: {df['perturbation'].max():.2f}")
print(f"Mean: {df['perturbation'].mean():.2f}")
print(f"Std: {df['perturbation'].std():.2f}")
# Determine components to analyze
if args.treatments:
components_to_analyze = args.treatments
else:
# Default to top-k components
components_to_analyze = list_available_components(df)[:args.top_k]
print(f"\nAnalyzing {len(components_to_analyze)} components as treatments: {', '.join(components_to_analyze)}")
# Run DoWhy causal analysis for each treatment component
results = analyze_components_with_dowhy(df, components_to_analyze)
# Save results to JSON file
output_filename = 'dowhy_causal_effects.json'
if args.test:
output_filename = 'test_dowhy_causal_effects.json'
output_path = os.path.join(os.path.dirname(__file__), output_filename)
try:
with open(output_path, 'w') as f:
json.dump({
"metadata": {
"json_file": json_file,
"test_mode": args.test,
"components_analyzed": components_to_analyze,
},
"results": results
}, f, indent=2)
logger.info(f"Causal analysis results saved to {output_path}")
print(f"\nCausal analysis complete. Results saved to {output_path}")
except Exception as e:
logger.error(f"Error saving results to {output_path}: {str(e)}")
print(f"\nError saving results: {str(e)}")
if __name__ == "__main__":
main() |