Getting training iterations and parameters of the model

, , ,
1

When I use prodigy train --ner to train to a model, it returned a model with some values and score. How can I get to know the no of iterations and other parameters to train the model?

2

Hi @ta13, the iterations are shown in the output after training. However, the number of iterations is not often "interesting" because the train command is configured to stop automatically when the model stops improving.

If you wish to control how long the training progresses, you can check the training configuration. You'd probably want to play around patience, epochs, etc.

3

How can I use Spacy's config file in Prodigy? Can you please provide me an example? I normally use prodigy's train command: prodigy train --ner . Now if I want to change epochs and patience, how can I use spacy's config in prodigy whereas prodigy also has prodigy.json file.

4

Hi @ta13,

You can either:

  1. Use the --config or -c argument to provide the path to the config.cfg file. OR
  2. Override it through the command-line interface. For example, if you with to override the patience parameter, you can pass --training.patience in the command line.
1 Like