configs/preset/drug_vqa.yaml

# @package _global_
defaults:
  - override /data/drug_featurizer: tokenizer
  - override /data/protein_featurizer: label
  - override /model/predictor: drug_vqa

model:
  loss:
    _target_: deepscreen.models.loss.multitask_loss.MultitaskWeightedLoss
    loss_fns:
      - ${task.loss}
      - _target_: deepscreen.models.predictors.drug_vqa.AttentionL2Regularization
    weights: [1, 0.001]

data:
  batch_size: 512
  drug_featurizer:
    tokenizer:
      _target_: deepscreen.data.featurizers.token.SmilesTokenizer
      vocab_file: resources/vocabs/drug_vqa/combinedVoc-wholeFour.voc
      regex_pattern: '(\[[^\[\]]{1,10}\])'