Exclusive categories not working as expected

, ,
1

Hi, I have a data set of recipes. I'm trying to add mutually exclusive categories to each recipe. To start, I tried to identify just desserts.

When I ran that experiment then tested the model in Spacy, I got really great results.

Now, I'm trying the experiment again. This time by categorizing recipes as Desserts or Cocktails. When I run the model against test results I see that each entry has both Desserts and Cocktails.

I see some references to an --exclusive flag in testcat.batch-train but I'm using Prodigy 1.9 and was directed by the application to use "train testcat." That command doesn't have an -E flag.

My workflow looks like this:

  1. build a seed list for desserts following your example in the Insults video
  2. export that seeds list as patterns file.
  3. build a seed list for cocktails
  4. export that seeds list as a patterns file
  5. merge the two patterns file
  6. Textcat.teach against a sample dataset using the merged patterns file
  7. Train the model.
  8. Run spacy test code against a second data set.

An example output from my test code:
slow cooker ground beef stroganoff slow cooker gro,{'DESSERTS': 0.0001183362037409097, 'COCKTAILS': 4.539786823443137e-05}

So, a couple of questions.

  1. What should I be doing differently?
  2. Is there a more efficient workflow?
2

If you're using prodigy train textcat, you should be able to set the flag -TE or --textcat-exclusive. You can see the full list of options when running prodigy train --help.

Let me know if that works!

3

That seems to work for me.