This is probably an easy one… if I omit the --output parameter from ner.batch-train (e.g. ‘ner.batch-train toptech_01 en_core_web_lg’), will it train the input model “in place”? Thanks!
No, if you leave out the
--output, it will discard the model that was trained and kept in memory, and only show you the results.
In most cases, you expect the results to be reproducable anyways, and instead of storing all those intermediate models you produce in your experiments, you often just want the figures to get an idea how the model performs, and only save it out once you’re actually ready to use it.
Thanks, Ines – that makes sense as a way to check for improvement between annotation sessions