config.yaml

global_config:
  seed: 5959
  device: "cuda"
  is_mp: True
  output_model_path: "next-visit-model"
  pretrained_model_path: "FL-mimic-iv-icd10-with-aggregations-mlm-model"  # pretrained MLM path

---
data_config:
  dataset_name: "NextVisit-MIMIC-IV"
  data_dir_path: "multi_center_dir_path"
  test_path: "test_mimic_iv_behrt_with_aggregations_next_visit_ds.csv"
  vocab_pickle_path: "mimic_iv_icd10_with_aggregations_vocab"
  max_patient_age: 110
  max_len_seq: 64
  min_visit: 1
---

fed_config:
  C: 0.1
  R: 500
  E: 10
  B: 8
---
init_config:
  init_type: "xavier"
  init_gain: 1.0
  gpu_ids: [ 0 ]
---
model_config:
  name: CustomBertForMultiLabelPrediction
  hidden_size: 288  # word embedding and seg embedding hidden size
  seg_vocab_size: 2  # number of vocab for seg embedding
  max_position_embedding: 64  # maximum number of tokens
  hidden_dropout_prob: 0.1  # dropout rate
  num_hidden_layers: 6  # number of multi-head attention layers required
  num_attention_heads: 12  # number of attention heads
  attention_probs_dropout_prob: 0.1  # multi-head attention dropout rate
  intermediate_size: 512  # the size of the "intermediate" layer in the transformer encoder
  hidden_act: gelu
  # The non-linear activation function in the encoder and the pooler "gelu", relu, swish are supported
  initializer_range: 0.02  # parameter weight initializer range
  feature_dict_config:
    word: true
    seg: true
    age: true
    position: true

---
log_config:
  log_path: "./log/"
  log_name: "FL.log"
  tb_port: 9989
  tb_host: "0.0.0.0"