Spaces:
Running
Running
File size: 12,143 Bytes
c185c85 |
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 |
import torch
import numpy as np
import logging
import plotly.graph_objects as go
from typing import Tuple, Dict
# Advanced analysis imports
import shap
import lime
from lime.lime_text import LimeTextExplainer
from config import config
from models import ModelManager, handle_errors
logger = logging.getLogger(__name__)
class AdvancedAnalysisEngine:
"""Advanced analysis using SHAP and LIME with FIXED implementation"""
def __init__(self):
self.model_manager = ModelManager()
def create_prediction_function(self, model, tokenizer, device):
"""Create FIXED prediction function for SHAP/LIME"""
def predict_proba(texts):
# Ensure texts is a list
if isinstance(texts, str):
texts = [texts]
elif isinstance(texts, np.ndarray):
texts = texts.tolist()
# Convert all elements to strings
texts = [str(text) for text in texts]
results = []
batch_size = 16 # Process in smaller batches
for i in range(0, len(texts), batch_size):
batch_texts = texts[i:i + batch_size]
try:
with torch.no_grad():
# Tokenize batch
inputs = tokenizer(
batch_texts,
return_tensors="pt",
padding=True,
truncation=True,
max_length=config.MAX_TEXT_LENGTH
).to(device)
# Batch inference
outputs = model(**inputs)
probs = torch.nn.functional.softmax(outputs.logits, dim=-1).cpu().numpy()
results.extend(probs)
except Exception as e:
logger.error(f"Prediction batch failed: {e}")
# Return neutral predictions for failed batch
batch_size_actual = len(batch_texts)
if hasattr(model.config, 'num_labels') and model.config.num_labels == 3:
neutral_probs = np.array([[0.33, 0.34, 0.33]] * batch_size_actual)
else:
neutral_probs = np.array([[0.5, 0.5]] * batch_size_actual)
results.extend(neutral_probs)
return np.array(results)
return predict_proba
@handle_errors(default_return=("Analysis failed", None, None))
def analyze_with_shap(self, text: str, language: str = 'auto', num_samples: int = 100) -> Tuple[str, go.Figure, Dict]:
"""FIXED SHAP analysis implementation"""
if not text.strip():
return "Please enter text for analysis", None, {}
# Detect language and get model
if language == 'auto':
detected_lang = self.model_manager.detect_language(text)
else:
detected_lang = language
model, tokenizer = self.model_manager.get_model(detected_lang)
try:
# Create FIXED prediction function
predict_fn = self.create_prediction_function(model, tokenizer, self.model_manager.device)
# Test the prediction function first
test_pred = predict_fn([text])
if test_pred is None or len(test_pred) == 0:
return "Prediction function test failed", None, {}
# Use SHAP Text Explainer instead of generic Explainer
explainer = shap.Explainer(predict_fn, masker=shap.maskers.Text(tokenizer))
# Get SHAP values with proper text input
shap_values = explainer([text], max_evals=num_samples)
# Extract data safely
if hasattr(shap_values, 'data') and hasattr(shap_values, 'values'):
tokens = shap_values.data[0] if len(shap_values.data) > 0 else []
values = shap_values.values[0] if len(shap_values.values) > 0 else []
else:
return "SHAP values extraction failed", None, {}
if len(tokens) == 0 or len(values) == 0:
return "No tokens or values extracted from SHAP", None, {}
# Handle multi-dimensional values
if len(values.shape) > 1:
# Use positive class values (last column for 3-class, second for 2-class)
pos_values = values[:, -1] if values.shape[1] >= 2 else values[:, 0]
else:
pos_values = values
# Ensure we have matching lengths
min_len = min(len(tokens), len(pos_values))
tokens = tokens[:min_len]
pos_values = pos_values[:min_len]
# Create visualization
fig = go.Figure()
colors = ['red' if v < 0 else 'green' for v in pos_values]
fig.add_trace(go.Bar(
x=list(range(len(tokens))),
y=pos_values,
text=tokens,
textposition='outside',
marker_color=colors,
name='SHAP Values',
hovertemplate='<b>%{text}</b><br>SHAP Value: %{y:.4f}<extra></extra>'
))
fig.update_layout(
title=f"SHAP Analysis - Token Importance (Samples: {num_samples})",
xaxis_title="Token Index",
yaxis_title="SHAP Value",
height=500,
xaxis=dict(tickmode='array', tickvals=list(range(len(tokens))), ticktext=tokens)
)
# Create analysis summary
analysis_data = {
'method': 'SHAP',
'language': detected_lang,
'total_tokens': len(tokens),
'samples_used': num_samples,
'positive_influence': sum(1 for v in pos_values if v > 0),
'negative_influence': sum(1 for v in pos_values if v < 0),
'most_important_tokens': [(str(tokens[i]), float(pos_values[i]))
for i in np.argsort(np.abs(pos_values))[-5:]]
}
summary_text = f"""
**SHAP Analysis Results:**
- **Language:** {detected_lang.upper()}
- **Total Tokens:** {analysis_data['total_tokens']}
- **Samples Used:** {num_samples}
- **Positive Influence Tokens:** {analysis_data['positive_influence']}
- **Negative Influence Tokens:** {analysis_data['negative_influence']}
- **Most Important Tokens:** {', '.join([f"{token}({score:.3f})" for token, score in analysis_data['most_important_tokens']])}
- **Status:** SHAP analysis completed successfully
"""
return summary_text, fig, analysis_data
except Exception as e:
logger.error(f"SHAP analysis failed: {e}")
error_msg = f"""
**SHAP Analysis Failed:**
- **Error:** {str(e)}
- **Language:** {detected_lang.upper()}
- **Suggestion:** Try with a shorter text or reduce number of samples
**Common fixes:**
- Reduce sample size to 50-100
- Use shorter input text (< 200 words)
- Check if model supports the text language
"""
return error_msg, None, {}
@handle_errors(default_return=("Analysis failed", None, None))
def analyze_with_lime(self, text: str, language: str = 'auto', num_samples: int = 100) -> Tuple[str, go.Figure, Dict]:
"""FIXED LIME analysis implementation - Bug Fix for mode parameter"""
if not text.strip():
return "Please enter text for analysis", None, {}
# Detect language and get model
if language == 'auto':
detected_lang = self.model_manager.detect_language(text)
else:
detected_lang = language
model, tokenizer = self.model_manager.get_model(detected_lang)
try:
# Create FIXED prediction function
predict_fn = self.create_prediction_function(model, tokenizer, self.model_manager.device)
# Test the prediction function first
test_pred = predict_fn([text])
if test_pred is None or len(test_pred) == 0:
return "Prediction function test failed", None, {}
# Determine class names based on model output
num_classes = test_pred.shape[1] if len(test_pred.shape) > 1 else 2
if num_classes == 3:
class_names = ['Negative', 'Neutral', 'Positive']
else:
class_names = ['Negative', 'Positive']
# Initialize LIME explainer - FIXED: Remove 'mode' parameter
explainer = LimeTextExplainer(class_names=class_names)
# Get LIME explanation
exp = explainer.explain_instance(
text,
predict_fn,
num_features=min(20, len(text.split())), # Limit features
num_samples=num_samples
)
# Extract feature importance
lime_data = exp.as_list()
if not lime_data:
return "No LIME features extracted", None, {}
# Create visualization
words = [item[0] for item in lime_data]
scores = [item[1] for item in lime_data]
fig = go.Figure()
colors = ['red' if s < 0 else 'green' for s in scores]
fig.add_trace(go.Bar(
y=words,
x=scores,
orientation='h',
marker_color=colors,
text=[f'{s:.3f}' for s in scores],
textposition='auto',
name='LIME Importance',
hovertemplate='<b>%{y}</b><br>Importance: %{x:.4f}<extra></extra>'
))
fig.update_layout(
title=f"LIME Analysis - Feature Importance (Samples: {num_samples})",
xaxis_title="Importance Score",
yaxis_title="Words/Phrases",
height=500
)
# Create analysis summary
analysis_data = {
'method': 'LIME',
'language': detected_lang,
'features_analyzed': len(lime_data),
'samples_used': num_samples,
'positive_features': sum(1 for _, score in lime_data if score > 0),
'negative_features': sum(1 for _, score in lime_data if score < 0),
'feature_importance': lime_data
}
summary_text = f"""
**LIME Analysis Results:**
- **Language:** {detected_lang.upper()}
- **Features Analyzed:** {analysis_data['features_analyzed']}
- **Classes:** {', '.join(class_names)}
- **Samples Used:** {num_samples}
- **Positive Features:** {analysis_data['positive_features']}
- **Negative Features:** {analysis_data['negative_features']}
- **Top Features:** {', '.join([f"{word}({score:.3f})" for word, score in lime_data[:5]])}
- **Status:** LIME analysis completed successfully
"""
return summary_text, fig, analysis_data
except Exception as e:
logger.error(f"LIME analysis failed: {e}")
error_msg = f"""
**LIME Analysis Failed:**
- **Error:** {str(e)}
- **Language:** {detected_lang.upper()}
- **Suggestion:** Try with a shorter text or reduce number of samples
**Bug Fix Applied:**
- ✅ Removed 'mode' parameter from LimeTextExplainer initialization
- ✅ This should resolve the "unexpected keyword argument 'mode'" error
**Common fixes:**
- Reduce sample size to 50-100
- Use shorter input text (< 200 words)
- Check if model supports the text language
"""
return error_msg, None, {} |