Config:
SVM: classifier_config:
- type: "SVM"
- top_k
- classification_type train_config:
- n_jobs
- max_iter
- optim_avg
- min_df
- max_df
- loss
- c
TRANSFORMER: classifier_config:
- type: "TRANSFORMER"
- top_k
- classification_type train_config:
- num_epochs
- seed
- optim_avg
- batch_size
- eval_batch_size
- grad_acc
- learning_rate
- weight_decay
- max_seq_len
- patience
Todo:
-
Return label names
-
CLI script
-
warning if inferring multilabel on trained as multiclass and viceversa. warning when training multilabel on multiclass dataset and viceversa.
-
which metric to optimize? micro-f, macro-f, weighted-f... parametrizable?
-
add logging
-
Transformer: predict_probabilities: improve and refactor
-
Transformer: dev, compute_metrics, evaluate_logits
-
Transformer: hyperparameter_search
-
SVM: test hp_space
-
package, versions and make pip installable
-
Notebook(s)