The --factor
option of ner.batch-train
appears to be ignored.
$ pgy ner.batch-train dataset en --output-model model --label LABEL --factor 1
Using 1 labels: LABEL
Loaded model en
Using 50% of accept/reject examples (132) for evaluation
Using 100% of remaining examples (1838) for training
Dropout: 0.2 Batch size: 32 Iterations: 10
BEFORE 0.000
Correct 0
Incorrect 82
Entities 3872
Unknown 1
# LOSS RIGHT WRONG ENTS SKIP ACCURACY
0%| | 0/1838 [00:00<?, ?it/s]
I would expect that with --factor 1
Prodigy would use all the data for training and none for validation.
This is version 1.4.0.