config_e2e_MLP_train.yaml

experiments_dir: ${HOME}/projects/E2EChallenge/experiments
log_level: INFO
random_seed: 1

# choose from: predict, train
mode: train

# If mode is 'predict', specify the model name,
# e.g. /path/to/model/weights.epoch5
model_fn:

# Model module names
data-module: components.data.e2e_data_MLP
model-module: components.model.e2e_model_MLP
training-module: components.trainer.e2e_trainer_MLP
evaluation-module: components.evaluator.e2e_evaluator

data_params:

  train_data: ${HOME}/var/E2E_challenge/trainset.csv
  dev_data: ${HOME}/var/E2E_challenge/devset.csv
  test_data:

  max_src_len: 50
  max_tgt_len: 50

model_params:

  embedding_dim: 256
  embedding_dropout: 0.1
  teacher_forcing_ratio: 1.0

  encoder_params:
    input_size: 256 # NOTE: should be equal to embedding_dim
    hidden_size: 512
    dropout: 0.0

  decoder_params:
    input_size: 512
    hidden_size: 512
    dropout: 0.0

training_params:
  evaluate_prediction: True
  save_model_each_epoch: True
  n_epochs: 30
  batch_size: 16
  optimizer: SGD
  learning_rate: 0.1