Feat/dl add emotion detection enhancements

configs/samo_emotion_detection_config.yaml

@@ -0,0 +1,186 @@
+# SAMO-DL Emotion Detection Configuration
+# Optimized parameters for journal entry emotion analysis
+
+# Model Configuration
+model:
+  name: "bert-base-uncased"  # Robust BERT model for emotion understanding
+  device: null               # Auto-detect (CPU/GPU)
+
+# Emotion Detection Parameters
+emotion_detection:
+  # Number of emotion categories (27 GoEmotions + neutral)
+  num_emotions: 28
+
+  # Prediction threshold for binary classification
+  prediction_threshold: 0.6  # Updated from 0.5 for better calibration
+
+  # Temperature scaling for calibrated predictions
+  temperature: 1.0
+
+  # Top-k emotions to return per prediction
+  top_k: 5
+
+# Model Architecture
+architecture:
+  # BERT configuration
+  hidden_dropout_prob: 0.3      # Dropout for BERT hidden layers
+  attention_probs_dropout_prob: 0.3  # Dropout for attention layers
+
+  # Classification head configuration
+  classifier_dropout_prob: 0.5  # Dropout for classification layers
+
+  # Freezing strategy
+  freeze_bert_layers: 6         # Number of BERT layers to freeze initially
+
+  # Class balancing
+  use_class_weights: true       # Enable class weight balancing
+
+# Training Configuration
+training:
+  # Batch sizes
+  train_batch_size: 16
+  eval_batch_size: 32
+
+  # Learning rates
+  bert_learning_rate: 2e-5      # Lower LR for BERT (fine-tuning)
+  classifier_learning_rate: 5e-4  # Higher LR for classification head
+
+  # Training epochs
+  num_epochs: 10
+  warmup_steps: 100
+
+  # Gradient settings
+  max_grad_norm: 1.0
+  gradient_accumulation_steps: 1
+
+  # Early stopping
+  early_stopping_patience: 3
+  early_stopping_threshold: 0.01
+
+# Data Processing
+data:
+  # Text processing
+  max_length: 512               # Maximum sequence length
+  truncation: true
+  padding: "max_length"
+
+  # Data augmentation
+  enable_augmentation: false    # Disable for now, can be enabled later
+
+  # Validation split
+  validation_split: 0.2
+  test_split: 0.1
+
+# Evaluation Configuration
+evaluation:
+  # Metrics to compute
+  metrics:
+    - "precision"
+    - "recall"
+    - "f1_micro"
+    - "f1_macro"
+    - "accuracy"
+
+  # Evaluation threshold (lowered to capture more predictions)
+  threshold: 0.2
+
+  # Top-k evaluation
+  top_k_evaluation: true
+  top_k_values: [1, 3, 5]
+
+# Logging and Monitoring
+logging:
+  level: "INFO"
+  log_interval: 100             # Log every N steps
+  save_interval: 1000           # Save checkpoint every N steps
+
+  # TensorBoard logging
+  enable_tensorboard: true
+  log_dir: "logs/emotion_detection"
+
+# Model Saving
+model_saving:
+  # Save directory
+  save_dir: "models/emotion_detection"
+
+  # Save best model based on metric
+  save_best_metric: "f1_macro"
+
+  # Save checkpoints
+  save_checkpoints: true
+  checkpoint_interval: 1        # Save every N epochs
+
+# Performance Optimization
+performance:
+  # Mixed precision training
+  use_amp: true                 # Automatic Mixed Precision
+
+  # Data loading
+  num_workers: 4
+  pin_memory: true
+
+  # Memory optimization
+  gradient_checkpointing: false  # Can be enabled for memory savings
+
+  # Inference optimization
+  use_torchscript: false        # Can be enabled for faster inference
+
+# SAMO-Specific Optimizations
+samo_optimizations:
+  # Journal entry specific settings
+  journal_entry_mode: true
+
+  # Emotional context awareness
+  context_awareness: true
+
+  # Multi-label prediction
+  multi_label_mode: true
+
+  # Confidence calibration
+  calibration_enabled: true
+
+  # Emotion intensity scaling
+  intensity_scaling: true
+
+# Error Handling
+error_handling:
+  # Retry settings
+  max_retries: 3
+  retry_delay: 1.0
+
+  # Fallback behavior
+  fallback_to_cpu: true
+  graceful_degradation: true
+
+  # Logging errors
+  log_errors: true
+  error_log_file: "logs/emotion_detection_errors.log"
+
+# Security and Privacy
+security:
+  # Input sanitization
+  sanitize_input: true
+
+  # Output filtering
+  filter_sensitive_emotions: false  # Can be enabled for privacy
+
+  # Rate limiting
+  rate_limit_requests: 1000    # Requests per minute
+
+  # Data privacy
+  anonymize_predictions: false  # Can be enabled for privacy
+
+# Development and Debugging
+development:
+  # Debug mode
+  debug_mode: false
+
+  # Verbose logging
+  verbose: false
+
+  # Test mode
+  test_mode: false
+
+  # Profiling
+  enable_profiling: false
+  profile_steps: 100

src/models/emotion_detection/config.py

@@ -0,0 +1,135 @@
+#!/usr/bin/env python3
+"""
+Configuration module for SAMO Emotion Detection System
+
+This module provides centralized configuration management for the emotion
+detection system, ensuring consistency across training and evaluation.
+"""
+
+from typing import Dict, Any
+from dataclasses import dataclass
+
+# Default configuration values
+DEFAULT_CONFIG = {
+    "model": {
+        "name": "bert-base-uncased",
+        "num_emotions": 28,
+        "hidden_dropout_prob": 0.3,
+        "classifier_dropout_prob": 0.5,
+        "freeze_bert_layers": 6,
+        "temperature": 1.0,
+    },
+    "training": {
+        "batch_size": 16,
+        "learning_rate": 2e-5,
+        "num_epochs": 10,
+        "weight_decay": 0.01,
+    },
+    "evaluation": {
+        "threshold": 0.2,  # Default evaluation threshold
+        "top_k": 5,
+    },
+    "prediction": {
+        "threshold": 0.6,  # Default prediction threshold
+        "max_length": 512,
+    }
+}
+
+
+def get_evaluation_threshold() -> float:
+    """Get current evaluation threshold from global config."""
+    return _config.evaluation_threshold
+
+
+def get_prediction_threshold() -> float:
+    """Get current prediction threshold from global config."""
+    return _config.prediction_threshold
+
+
+@dataclass
+class EmotionDetectionConfig:
+    """Configuration class for emotion detection system."""
+
+    # Model configuration
+    model_name: str = "bert-base-uncased"
+    num_emotions: int = 28
+    hidden_dropout_prob: float = 0.3
+    classifier_dropout_prob: float = 0.5
+    freeze_bert_layers: int = 6
+    temperature: float = 1.0
+
+    # Training configuration
+    batch_size: int = 16
+    learning_rate: float = 2e-5
+    num_epochs: int = 10
+    weight_decay: float = 0.01
+
+    # Evaluation configuration
+    evaluation_threshold: float = 0.2
+    top_k: int = 5
+
+    # Prediction configuration
+    prediction_threshold: float = 0.6
+    max_length: int = 512
+
+    @classmethod
+    def from_dict(cls, config_dict: Dict[str, Any]) -> 'EmotionDetectionConfig':
+        """Create config from dictionary."""
+        return cls(**config_dict)
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert config to dictionary."""
+        return {
+            "model_name": self.model_name,
+            "num_emotions": self.num_emotions,
+            "hidden_dropout_prob": self.hidden_dropout_prob,
+            "classifier_dropout_prob": self.classifier_dropout_prob,
+            "freeze_bert_layers": self.freeze_bert_layers,
+            "temperature": self.temperature,
+            "batch_size": self.batch_size,
+            "learning_rate": self.learning_rate,
+            "num_epochs": self.num_epochs,
+            "weight_decay": self.weight_decay,
+            "evaluation_threshold": self.evaluation_threshold,
+            "top_k": self.top_k,
+            "prediction_threshold": self.prediction_threshold,
+            "max_length": self.max_length,
+        }
+
+
+def get_default_config() -> EmotionDetectionConfig:
+    """Get default configuration."""
+    return EmotionDetectionConfig()
+
+
+def get_config_from_dict(config_dict: Dict[str, Any]) -> EmotionDetectionConfig:
+    """Get configuration from dictionary with defaults."""
+    default_config = get_default_config()
+
+    # Update with provided values
+    for key, value in config_dict.items():
+        if hasattr(default_config, key):
+            setattr(default_config, key, value)
+
+    return default_config
+
+
+# Global configuration instance
+_config = get_default_config()
+
+
+def get_config() -> EmotionDetectionConfig:
+    """Get current configuration."""
+    return _config
+
+
+def update_config(config_dict: Dict[str, Any]) -> None:
+    """Update global configuration."""
+    global _config
+    _config = get_config_from_dict(config_dict)
+
+
+def reset_config() -> None:
+    """Reset to default configuration."""
+    global _config
+    _config = get_default_config()