Ensemble Models

[aditya0by0]

• Implementation for Add ensemble model #72

[aditya0by0]

Test example config :

class_path: chebai.models.ensemble.ChebiEnsemble
init_args:
  optimizer_kwargs:
    lr: 1e-3
  model_configs: {
    "chebi50": {
      "ckpt_path": "logs/chebi50_bce_unweighted/version_9/checkpoints/best_epoch=00_val_loss=0.3521_val_macro-f1=0.0381_val_micro-f1=0.1753.ckpt",
      "class_path": chebai.models.Electra,
      "labels_path": "data/chebi_v231/ChEBI50/processed/classes.json",
    },

    "chebi50_1": {
      "ckpt_path": "logs/chebi50_bce_unweighted/version_9/checkpoints/best_epoch=00_val_loss=0.3521_val_macro-f1=0.0381_val_micro-f1=0.1753.ckpt",
      "class_path": chebai.models.Electra,
      "labels_path": "data/chebi_v231/ChEBI50/processed/classes.json",
    },
  }

[sfluegel05]

Why does the ensemble model need a learning rate? I would assume that it cannot be trained at all and is a prediction-only model