seqtagger/params/seqtag_params.yaml

corpus:
  columns: # columns definition for ColumnCorpus
    0: 'text'
    1: 'ner'
    2: 'space-after'
  data_folder: "./data/"
  train_file: "train.txt"
  test_file: "test.txt"
  dev_file: "dev.txt"

embeddings:
  flair-backward: "./models/best-flair-backward.pt"
  fasttext: "./models/fasttext_jurinet_jurica_01012020_01072020_anonym.gensim"

seqtagger:
  hidden_size: 32 # hidden_size: number of hidden states in RNN
  rnn_layers: 1 # rnn_layers: number of RNN layers
  use_crf: False # use_crf: if True use CRF decoder, else project directly to tag space
  dropout: 0.0 # dropout: dropout probability
  word_dropout: 0.05 # word_dropout: word dropout probability
  reproject_embeddings: True # reproject_embeddings: if True, adds trainable linear map on top of embedding layer. If False, no map.
  train_initial_hidden_state: False # train_initial_hidden_state: if True, trains initial hidden state of RNN

trainer:
  epoch: 8 # epoch: The starting epoch (normally 0 but could be higher if you continue training model)
  use_tensorboard: True # use_tensorboard: If True, writes out tensorboard information
  #optimizer: "SGD" # optimizer: The optimizer to use (typically SGD or Adam)

type:
  is_fresh_start : False
  run_name: "run3"

train:
  base_path: "./outputs/models/vanilla/" # base_path: Main path to which all output during training is logged and models are saved
  learning_rate: 0.01 # learning_rate: Initial learning rate
  mini_batch_size: 5 # mini_batch_size: Size of mini-batches during training
  max_epochs: 16 # max_epochs: Maximum number of epochs to train. Terminates training if this number is surpassed
  anneal_factor: 0.5 # anneal_factor: The factor by which the learning rate is annealed
  embeddings_storage_mode: "cpu" # embeddings_storage_mode: One of 'none' (all embeddings are deleted and freshly recomputed) 
  patience: 2 # patience: Patience is the number of epochs with no improvement the Trainer waits until annealing the learning rate
  min_learning_rate: 0.1 # min_learning_rate: If the learning rate falls below this threshold, training terminates
  train_with_dev: False # train_with_dev: If True, training is performed using both train+dev data
  monitor_train: False # monitor_train: If True, training data is evaluated at end of each epoch
  monitor_test: True # monitor_test: If True, test data is evaluated at end of each epoch
  checkpoint: False # checkpoint: If True, a full checkpoint is saved at end of each epoch
  save_final_model: True  # save_final_model: If True, final model is saved
  shuffle: True # shuffle: If True, data is shuffled during training