`output_dir` Specification Error

,
1

according to the documentation i should be able to set an output directory for my model using

 prodigy train ./output_models --textcat-multilabel mydata --label-stats True --base-model en_core_web_lg --lang "en" --eval-split 0.2

This returns an error:

✘ Invalid config override 'True': name should start with --

According to the command line help:

$  prodigy train --help
usage: prodigy train [-h] [-n ] [-tc ] [-tcm ] [-t ] [-s ] [-p ] [-sc ] [-NM] [-es 0.2] [-LS] [-c None] [-m None] [-l en] [-g -1] [-V] [-S]
                     [output_dir] [overrides ...]

Train a spaCy pipeline from one or more datasets for different components.
    Per-component evaluation sets can be provided using the eval: prefix, e.g.
    --ner my_train_dataset,eval:my_eval_dataset. The command takes care of
    auto-generating a training config and merging all annotations on the
    same input data.
    
positional arguments:
  output_dir            Optional output directory for the trained pipeline
  overrides             Config overrides

yet, when i run this command,

 prodigy train --textcat-multilabel mydata --label-stats True --base-model en_core_web_lg --lang "en" --eval-split 0.2 ./output_models

I also get an error:

✘ Invalid config override './output_models': name should start with
--

Is there a problem with the documentation or is this a problem with the way I am specifying the directory?

If so, how can I specify an output directory?

2

Hi! It looks like the problem here might be this:

If you're setting boolean flag on the CLI, you don't have to actually specify the value – just setting --label-stats will evaluate to True under the hood. So I think the CLI argument parsing gets confused here by the boolean flag being followed by a value that doesn't start with --.