config/train_config_augmented.yaml

# DCE Intent BioClinicalBERT Training Configuration - Augmented Dataset
# ======================================================================
# Version: 2.1.0
# Last Updated: 2026-02-13
# Description: Training config using merged dataset with disease symptom data

# Model Configuration
model:
  base_model: "emilyalsentzer/Bio_ClinicalBERT"
  max_length: 48
  num_labels: null  # Auto-detected from dataset

# Training Hyperparameters
# Slightly adjusted for larger dataset
training:
  learning_rate: 2.5e-5  # Slightly lower LR for larger dataset
  epochs: 12  # Slightly fewer epochs since more data
  batch_size: 16
  eval_batch_size: 32
  warmup_ratio: 0.1
  weight_decay: 0.01
  gradient_accumulation_steps: 1
  max_grad_norm: 1.0
  
# Optimizer
optimizer:
  type: "adamw"
  betas: [0.9, 0.999]
  eps: 1.0e-8

# Scheduler
scheduler:
  type: "linear"

# Data Paths - Using merged augmented dataset
data:
  train_path: "data/merged/train.jsonl"
  val_path: "data/merged/val.jsonl"
  test_path: "data/merged/test.jsonl"
  text_column: "text"
  label_column: "label"

# Output Configuration
output:
  base_dir: "artifacts"
  pytorch_dir: "artifacts/pytorch"
  save_steps: 50
  eval_steps: 25
  logging_steps: 25
  save_total_limit: 2

# Reproducibility
seed: 42
deterministic: true

# Logging
logging:
  level: "INFO"
  log_to_file: true
  log_file: "artifacts/logs/training_augmented.log"

# Intent Schema
intent_schema:
  tier1:
    - "ESCALATION"
    - "APPOINTMENT"
    - "MEDICATION"
    - "SYMPTOM_CHECK"
    - "GENERAL_INQUIRY"
    - "BILLING"
    - "OTHER"

# Evaluation Metrics
metrics:
  primary: "macro_f1"
  track:
    - "accuracy"
    - "macro_f1"
    - "per_class_recall"
  escalation_label: "ESCALATION"