config.yaml

Process:
  Stream: False # setting to true will be an expensive operation, stream data in batches, defined below.
  # If set to false, process will read the input ticket query from [Data][input_query_path]
  Stream_batch: 500 # 500 new tickets daily. To be used only if True.

Logging:
  log_dir: "logging"
  log_path: "app.log"

Data:
  original_ticket: 'path/to/support_tickets.json'
  ticket_train: "data/data_training_data.json"        # 90% portion of the original dataset, usd for training, ingesting into vector db and evaluating.
  ticket_test: "data/data_stream_test.json"           # 10% portion of the original dataset

  training_data: "data/data.json"                     # processed training data for deep learning model
  input_query_path: "data/input_query.json"           # Single Incoming ticket to be processed, used by single_stream fucntion, if Stream set to False

  final_output: "data/queried_result.json"            # Output path Incoming ticket after processiung

Train:
  train_model: False # this will train the model from scratch, else, load the trained model provided.
  train_tokenizer: False # this will train the tokenizer from scratch, else, load the trained tokenizer provided.
  VOCAB_SIZE: 5000
  BATCH_SIZE: 64
  Sample_per_class: 12 # 12 samples per batch for class imbalance.
  learning_rate: 0.0001
  EPOCHS: 20
  MAX_LEN: 100
  EMBEDDING_SIZE: 100
  ENCODER_HIDDEN_DIMENSION: 64

Model:
  model_path: "model/support_ticket_model.pth"
  tokenizer_path: "model/tokenizer.json"

RAG:
  embedding_model: "all-MiniLM-L6-v2"
  vector_store: "data/vector_store"
  vectordb_name: "support_ticket_db"
  ingestion_batch_size: 5461
  score_threshold: 0.65