File size: 8,581 Bytes
25d0747 | 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 | import torch
import torch.nn as nn
from torch.utils.data import DataLoader
import numpy as np
from sklearn.metrics import classification_report, confusion_matrix, f1_score
from sklearn.metrics import mean_absolute_error, mean_squared_error
import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd
import argparse
import os
from tqdm import tqdm
from models.vision import VisionEmotionModel
from models.audio import AudioEmotionModel
from models.text import TextIntentModel
from models.fusion import MultiModalFusion
def evaluate_model(model, dataloader, device, task='emotion'):
"""
Evaluate model on given task.
"""
model.eval()
all_preds = []
all_labels = []
with torch.no_grad():
for batch in tqdm(dataloader, desc=f"Evaluating {task}"):
if task == 'emotion':
vision = batch['vision'].to(device)
audio = batch['audio'].to(device)
text_input_ids = batch['text']['input_ids'].to(device)
text_attention_mask = batch['text']['attention_mask'].to(device)
labels = batch['emotion'].to(device)
outputs = model(vision, audio, text_input_ids, text_attention_mask)
preds = outputs['emotion'].argmax(dim=1)
elif task == 'intent':
# Similar for intent
preds = outputs['intent'].argmax(dim=1)
labels = batch['intent'].to(device)
all_preds.extend(preds.cpu().numpy())
all_labels.extend(labels.cpu().numpy())
return np.array(all_preds), np.array(all_labels)
def ablation_study(fusion_model, dataloader, device):
"""
Perform ablation study by removing modalities.
"""
print("Performing Ablation Study...")
results = {}
# Full model
preds, labels = evaluate_model(fusion_model, dataloader, device)
results['full'] = f1_score(labels, preds, average='weighted')
# Vision-only (set audio and text to zero)
fusion_model.eval()
ablation_preds = []
with torch.no_grad():
for batch in dataloader:
vision = batch['vision'].to(device)
audio = torch.zeros_like(batch['audio']).to(device)
text_input_ids = batch['text']['input_ids'].to(device)
text_attention_mask = batch['text']['attention_mask'].to(device)
outputs = fusion_model(vision, audio, text_input_ids, text_attention_mask)
preds = outputs['emotion'].argmax(dim=1)
ablation_preds.extend(preds.cpu().numpy())
results['vision_only'] = f1_score(labels, ablation_preds, average='weighted')
# Audio-only
ablation_preds = []
with torch.no_grad():
for batch in dataloader:
vision = torch.zeros_like(batch['vision']).to(device)
audio = batch['audio'].to(device)
text_input_ids = batch['text']['input_ids'].to(device)
text_attention_mask = batch['text']['attention_mask'].to(device)
outputs = fusion_model(vision, audio, text_input_ids, text_attention_mask)
preds = outputs['emotion'].argmax(dim=1)
ablation_preds.extend(preds.cpu().numpy())
results['audio_only'] = f1_score(labels, ablation_preds, average='weighted')
# Text-only
ablation_preds = []
with torch.no_grad():
for batch in dataloader:
vision = torch.zeros_like(batch['vision']).to(device)
audio = torch.zeros_like(batch['audio']).to(device)
text_input_ids = batch['text']['input_ids'].to(device)
text_attention_mask = batch['text']['attention_mask'].to(device)
outputs = fusion_model(vision, audio, text_input_ids, text_attention_mask)
preds = outputs['emotion'].argmax(dim=1)
ablation_preds.extend(preds.cpu().numpy())
results['text_only'] = f1_score(labels, ablation_preds, average='weighted')
return results
def bias_analysis(model, dataloader, device, demographic_groups):
"""
Analyze bias across demographic groups.
"""
print("Performing Bias Analysis...")
bias_results = {}
model.eval()
with torch.no_grad():
for group in demographic_groups:
group_preds = []
group_labels = []
# Filter data for this demographic group
# This would require demographic labels in dataset
for batch in dataloader:
# Placeholder: assume demographic info in batch
if 'demographic' in batch and batch['demographic'] == group:
vision = batch['vision'].to(device)
audio = batch['audio'].to(device)
text_input_ids = batch['text']['input_ids'].to(device)
text_attention_mask = batch['text']['attention_mask'].to(device)
outputs = model(vision, audio, text_input_ids, text_attention_mask)
preds = outputs['emotion'].argmax(dim=1)
labels = batch['emotion']
group_preds.extend(preds.cpu().numpy())
group_labels.extend(labels.cpu().numpy())
if group_preds:
bias_results[group] = {
'f1': f1_score(group_labels, group_preds, average='weighted'),
'accuracy': np.mean(np.array(group_preds) == np.array(group_labels))
}
return bias_results
def plot_confusion_matrix(cm, labels, save_path):
"""
Plot and save confusion matrix.
"""
plt.figure(figsize=(10, 8))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues',
xticklabels=labels, yticklabels=labels)
plt.title('Confusion Matrix')
plt.ylabel('True Label')
plt.xlabel('Predicted Label')
plt.tight_layout()
plt.savefig(save_path)
plt.close()
def generate_report(results, ablation_results, bias_results, output_dir):
"""
Generate comprehensive evaluation report.
"""
report = f"""
# EMOTIA Model Evaluation Report
## Overall Performance
- Emotion F1-Score: {results['emotion_f1']:.4f}
- Intent F1-Score: {results['intent_f1']:.4f}
- Engagement MAE: {results['engagement_mae']:.4f}
- Confidence MAE: {results['confidence_mae']:.4f}
## Ablation Study Results
{chr(10).join([f"- {k}: {v:.4f}" for k, v in ablation_results.items()])}
## Bias Analysis
"""
if bias_results:
for group, metrics in bias_results.items():
report += f"- {group}: F1={metrics['f1']:.4f}, Acc={metrics['accuracy']:.4f}\n"
else:
report += "No demographic data available for bias analysis.\n"
report += """
## Recommendations
- Focus on improving the weakest modality based on ablation results.
- Monitor and mitigate biases identified in demographic analysis.
- Consider additional data augmentation for underrepresented classes.
"""
with open(os.path.join(output_dir, 'evaluation_report.md'), 'w') as f:
f.write(report)
print("Evaluation report saved to evaluation_report.md")
def main(args):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Load model
fusion_model = MultiModalFusion().to(device)
fusion_model.load_state_dict(torch.load(args.model_path))
fusion_model.eval()
# Load test data
# test_dataset = MultiModalDataset(args.data_dir, 'test')
# test_loader = DataLoader(test_dataset, batch_size=args.batch_size)
# Placeholder for actual evaluation
print("Evaluation framework ready. Implement data loading for full evaluation.")
# Example results structure
results = {
'emotion_f1': 0.85,
'intent_f1': 0.78,
'engagement_mae': 0.12,
'confidence_mae': 0.15
}
ablation_results = {
'full': 0.85,
'vision_only': 0.72,
'audio_only': 0.68,
'text_only': 0.75
}
bias_results = {} # Would be populated with actual demographic analysis
# Generate report
generate_report(results, ablation_results, bias_results, args.output_dir)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Evaluate EMOTIA Model")
parser.add_argument('--model_path', type=str, required=True, help='Path to trained model')
parser.add_argument('--data_dir', type=str, required=True, help='Path to test data')
parser.add_argument('--output_dir', type=str, default='./evaluation_results', help='Output directory')
parser.add_argument('--batch_size', type=int, default=16, help='Batch size')
args = parser.parse_args()
main(args) |