Spaces:
Running
Running
File size: 14,658 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 |
#!/usr/bin/env python3
"""
Causal Component Analysis
This script implements causal inference methods 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 logging
import argparse
from typing import Dict, List, Optional, Tuple, Set
from sklearn.linear_model import LinearRegression
# Import from utils directory
from .utils.dataframe_builder import create_component_influence_dataframe
# Import shared utilities
from .utils.shared_utils import list_available_components
# Configure logging for this module
logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
def calculate_average_treatment_effect(
df: pd.DataFrame,
component_id: str,
outcome_var: str = "perturbation",
control_vars: Optional[List[str]] = None
) -> Dict[str, float]:
"""
Calculates the Average Treatment Effect (ATE) of a component on perturbation score.
Args:
df: DataFrame with binary component features and perturbation score
component_id: ID of the component to analyze (including 'entity_' or 'relation_' prefix)
outcome_var: Name of the outcome variable (default: 'perturbation')
control_vars: List of control variables to include in the model (other components)
Returns:
Dictionary with ATE estimates and confidence intervals
"""
if component_id not in df.columns:
logger.error(f"Component {component_id} not found in DataFrame")
return {
"ate": 0.0,
"std_error": 0.0,
"p_value": 1.0,
"confidence_interval_95": (0.0, 0.0)
}
# Check if there's enough variation in the treatment variable
if df[component_id].std() == 0:
logger.warning(f"No variation in component {component_id}, cannot estimate causal effect")
return {
"ate": 0.0,
"std_error": 0.0,
"p_value": 1.0,
"confidence_interval_95": (0.0, 0.0)
}
# Check if there's enough variation in the outcome variable
if df[outcome_var].std() == 0:
logger.warning(f"No variation in outcome {outcome_var}, cannot estimate causal effect")
return {
"ate": 0.0,
"std_error": 0.0,
"p_value": 1.0,
"confidence_interval_95": (0.0, 0.0)
}
# Select control variables (other components that could confound the relationship)
if control_vars is None:
# Use all other components as control variables
control_vars = [col for col in df.columns if (col.startswith("entity_") or col.startswith("relation_")) and col != component_id]
# Create treatment and control groups
treatment_group = df[df[component_id] == 1]
control_group = df[df[component_id] == 0]
# Calculate naive ATE (without controlling for confounders)
naive_ate = treatment_group[outcome_var].mean() - control_group[outcome_var].mean()
# Implement regression adjustment to control for confounders
X = df[control_vars + [component_id]]
y = df[outcome_var]
# Use linear regression for adjustment
model = LinearRegression()
model.fit(X, y)
# Extract coefficient for the component of interest (the ATE)
component_idx = control_vars.index(component_id) if component_id in control_vars else -1
ate = model.coef_[component_idx]
# Use bootstrapping to calculate standard errors and confidence intervals
# Simplified implementation for demonstration
n_bootstrap = 1000
bootstrap_ates = []
for _ in range(n_bootstrap):
# Sample with replacement
sample_idx = np.random.choice(len(df), len(df), replace=True)
sample_df = df.iloc[sample_idx]
# Calculate ATE for this sample
sample_X = sample_df[control_vars + [component_id]]
sample_y = sample_df[outcome_var]
try:
sample_model = LinearRegression()
sample_model.fit(sample_X, sample_y)
sample_ate = sample_model.coef_[component_idx]
bootstrap_ates.append(sample_ate)
except:
# Skip problematic samples
continue
# Calculate standard error and confidence intervals
std_error = np.std(bootstrap_ates)
ci_lower = np.percentile(bootstrap_ates, 2.5)
ci_upper = np.percentile(bootstrap_ates, 97.5)
# Calculate p-value (simplified approach)
z_score = ate / std_error if std_error > 0 else 0
p_value = 2 * (1 - abs(z_score)) if z_score != 0 else 1.0
return {
"ate": ate,
"naive_ate": naive_ate,
"std_error": std_error,
"p_value": p_value,
"confidence_interval_95": (ci_lower, ci_upper)
}
def granger_causality_test(
df: pd.DataFrame,
component_id: str,
outcome_var: str = "perturbation",
max_lag: int = 2
) -> Dict[str, float]:
"""
Implements a simplified Granger causality test to assess if a component
'Granger-causes' the perturbation score.
Note: This is a simplified implementation and requires time-series data.
If the data doesn't have a clear time dimension, the results should be
interpreted with caution.
Args:
df: DataFrame with binary component features and perturbation score
component_id: ID of the component to analyze (including 'entity_' or 'relation_' prefix)
outcome_var: Name of the outcome variable (default: 'perturbation')
max_lag: Maximum number of lags to include in the model
Returns:
Dictionary with Granger causality test results
"""
if component_id not in df.columns:
logger.error(f"Component {component_id} not found in DataFrame")
return {"f_statistic": 0.0, "p_value": 1.0, "causal_direction": "none"}
# Check if there's enough data points
if len(df) <= max_lag + 1:
logger.warning(f"Not enough data points for Granger causality test with max_lag={max_lag}")
return {"f_statistic": 0.0, "p_value": 1.0, "causal_direction": "none"}
# Check if there's enough variation in the variables
if df[component_id].std() == 0 or df[outcome_var].std() == 0:
logger.warning(f"No variation in component or outcome, cannot test Granger causality")
return {"f_statistic": 0.0, "p_value": 1.0, "causal_direction": "none"}
# Implement Granger causality test using OLS and F-test
# This is a simplified approach - in practice, use statsmodels or other libraries
# First, create lagged versions of the data
lagged_df = df.copy()
for i in range(1, max_lag + 1):
lagged_df[f"{component_id}_lag{i}"] = df[component_id].shift(i)
lagged_df[f"{outcome_var}_lag{i}"] = df[outcome_var].shift(i)
# Drop rows with NaN values (due to lagging)
lagged_df = lagged_df.dropna()
# Model 1: Outcome ~ Past Outcomes
X1 = lagged_df[[f"{outcome_var}_lag{i}" for i in range(1, max_lag + 1)]]
y = lagged_df[outcome_var]
model1 = LinearRegression()
model1.fit(X1, y)
y_pred1 = model1.predict(X1)
ssr1 = np.sum((y - y_pred1) ** 2)
# Model 2: Outcome ~ Past Outcomes + Past Component
X2 = lagged_df[[f"{outcome_var}_lag{i}" for i in range(1, max_lag + 1)] +
[f"{component_id}_lag{i}" for i in range(1, max_lag + 1)]]
model2 = LinearRegression()
model2.fit(X2, y)
y_pred2 = model2.predict(X2)
ssr2 = np.sum((y - y_pred2) ** 2)
# Calculate F-statistic
n = len(lagged_df)
df1 = max_lag
df2 = n - 2 * max_lag - 1
if ssr1 == 0 or df2 <= 0:
f_statistic = 0
p_value = 1.0
else:
f_statistic = ((ssr1 - ssr2) / df1) / (ssr2 / df2)
# Simplified p-value calculation (for demonstration)
p_value = 1 / (1 + f_statistic)
# Test reverse causality
# Model 3: Component ~ Past Components
X3 = lagged_df[[f"{component_id}_lag{i}" for i in range(1, max_lag + 1)]]
y_comp = lagged_df[component_id]
model3 = LinearRegression()
model3.fit(X3, y_comp)
y_pred3 = model3.predict(X3)
ssr3 = np.sum((y_comp - y_pred3) ** 2)
# Model 4: Component ~ Past Components + Past Outcomes
X4 = lagged_df[[f"{component_id}_lag{i}" for i in range(1, max_lag + 1)] +
[f"{outcome_var}_lag{i}" for i in range(1, max_lag + 1)]]
model4 = LinearRegression()
model4.fit(X4, y_comp)
y_pred4 = model4.predict(X4)
ssr4 = np.sum((y_comp - y_pred4) ** 2)
# Calculate F-statistic for reverse causality
if ssr3 == 0 or df2 <= 0:
f_statistic_reverse = 0
p_value_reverse = 1.0
else:
f_statistic_reverse = ((ssr3 - ssr4) / df1) / (ssr4 / df2)
# Simplified p-value calculation
p_value_reverse = 1 / (1 + f_statistic_reverse)
# Determine causality direction
causal_direction = "none"
if p_value < 0.05 and p_value_reverse >= 0.05:
causal_direction = "component -> outcome"
elif p_value >= 0.05 and p_value_reverse < 0.05:
causal_direction = "outcome -> component"
elif p_value < 0.05 and p_value_reverse < 0.05:
causal_direction = "bidirectional"
return {
"f_statistic": f_statistic,
"p_value": p_value,
"f_statistic_reverse": f_statistic_reverse,
"p_value_reverse": p_value_reverse,
"causal_direction": causal_direction
}
def compute_causal_effect_strength(
df: pd.DataFrame,
control_group: Optional[List[str]] = None,
outcome_var: str = "perturbation"
) -> Dict[str, float]:
"""
Computes the strength of causal effects for all components.
Args:
df: DataFrame with binary component features and perturbation score
control_group: List of components to use as control variables
outcome_var: Name of the outcome variable (default: 'perturbation')
Returns:
Dictionary mapping component IDs to their causal effect strengths
"""
# Get all component columns
component_cols = [col for col in df.columns if col.startswith(("entity_", "relation_"))]
if not component_cols:
logger.error("No component features found in DataFrame")
return {}
# Calculate ATE for each component
effect_strengths = {}
for component_id in component_cols:
try:
ate_results = calculate_average_treatment_effect(
df,
component_id,
outcome_var=outcome_var,
control_vars=control_group
)
effect_strengths[component_id] = ate_results["ate"]
except Exception as e:
logger.warning(f"Error calculating ATE for {component_id}: {e}")
effect_strengths[component_id] = 0.0
return effect_strengths
# Note: create_mock_perturbation_scores and list_available_components
# moved to utils.shared_utils to avoid duplication
def main():
"""Main function to run the causal component analysis."""
parser = argparse.ArgumentParser(description='Analyze causal relationships between components and perturbation scores')
parser.add_argument('--input', '-i', required=True, help='Path to the knowledge graph JSON file')
parser.add_argument('--output', '-o', help='Path to save the output analysis (CSV format)')
args = parser.parse_args()
print(f"Loading knowledge graph")
# Create DataFrame
df = create_component_influence_dataframe(args.input)
if df is None or df.empty:
logger.error("Failed to create or empty DataFrame. Cannot proceed with analysis.")
return
# Print basic DataFrame info
print(f"\nDataFrame info:")
print(f"Rows: {len(df)}")
entity_features = [col for col in df.columns if col.startswith("entity_")]
relation_features = [col for col in df.columns if col.startswith("relation_")]
print(f"Entity features: {len(entity_features)}")
print(f"Relation features: {len(relation_features)}")
# Check if we have any variance in perturbation scores
if df['perturbation'].std() == 0:
logger.warning("All perturbation scores are identical. This might lead to uninformative results.")
print("\nWARNING: All perturbation scores are identical (value: %.2f). Results may not be meaningful." % df['perturbation'].iloc[0])
else:
print(f"\nPerturbation score distribution:")
print(f"Min: {df['perturbation'].min():.2f}, Max: {df['perturbation'].max():.2f}")
print(f"Mean: {df['perturbation'].mean():.2f}, Std: {df['perturbation'].std():.2f}")
# Compute causal effect strengths
print("\nComputing causal effect strengths...")
effect_strengths = compute_causal_effect_strength(df)
print(f"Found {len(effect_strengths)} components with causal effects")
# Sort components by effect strength
sorted_components = sorted(effect_strengths.items(), key=lambda x: abs(x[1]), reverse=True)
print("\nTop 10 Components by Causal Effect Strength:")
print("=" * 50)
print(f"{'Rank':<5}{'Component':<30}{'Effect Strength':<15}")
print("-" * 50)
for i, (component, strength) in enumerate(sorted_components[:10], 1):
print(f"{i:<5}{component:<30}{strength:.6f}")
# Save results
if args.output:
# Create results DataFrame
results_df = pd.DataFrame({
'Component': [comp for comp, _ in sorted_components],
'Effect_Strength': [strength for _, strength in sorted_components]
})
# Save to specified output path
print(f"\nSaving results to: {args.output}")
try:
results_df.to_csv(args.output, index=False)
print(f"Successfully saved results to: {args.output}")
except Exception as e:
print(f"Error saving to {args.output}: {str(e)}")
# Also save to default location in the causal_analysis directory
default_output = os.path.join(os.path.dirname(__file__), 'causal_component_effects.csv')
print(f"Also saving results to: {default_output}")
try:
results_df.to_csv(default_output, index=False)
print(f"Successfully saved results to: {default_output}")
except Exception as e:
print(f"Error saving to {default_output}: {str(e)}")
print("\nAnalysis complete.")
if __name__ == "__main__":
main() |