Open inchiosa opened 4 years ago
Greetings @inchiosa -- I fully agree, this has been a rather long standing issue. We provide a way to automatically train a model from the commandline, but we do not provide a way to score a test dataset from the commandline.
I think it would look like:
mlnet predict --test-dataset TestData.tsv --model file.zip
This would do a bulk prediction of all rows in the test dataset.
One difficulty is knowing the input schema. For this I would recommend that the mlnet auto-train
stores the text-loader settings in the model zip file.
Provide a Predict CLI command, similar to the mlnet-autotrain command