ZeroShot Classfier (zs_classifier)

zs_classifier is a zeroshot text classification module build on top of flair library. It uses tars-base model for text classification

This tutorial shows you how to import and use zs_classifier with jac code.

1. Import ZS Classifier module in jac
2. ZeroShot Predictions
3. ZeroShot Embeddings

Walk through

1. Import ZeroShot Classfier(zs_classifier) module in jac

1. Open terminal and run jaseci by cmd

   jsctl -m

2. Load zs_classifier module in jac by cmd

   actions load module jac_nlp.zs_classifier

2. ZeroShot Predictions

• Creating a Jac Walker (predict_zs)

  1. Create a file by name zs_classifier.jac

  2. Create predict_zs walker

     walker predict_zs{
             can zs_classifier.classify;
             report zs_classifier.classify(text="I am so glad you liked it", classes=["happy","sad"]);
         }
     
     

     Parameter details

     • classify: used to classify text among the classes provided
       • Input:
         • text (str or List(str)): it can be string or list of string
         • classes (List): list of classes that text needs to be classified
       • Returns: list of classes with the confidence score for each

• Steps for running zs_classifier.jac program

  3. Build zs_classifier.jac by run cmd

     jac build zs_classifier.jac

  4. Activate sentinal by run cmd

     sentinel set -snt active:sentinel -mode ir zs_classifier.jir

  5. Calling walker predict_zs with default parameter

     walker run predict_zs
     

  Expected output

  [
      {
          "I am so glad you liked it": [
              {
                  "value": "happy",
                  "confidence": 0.8714166879653931
              }
          ]
      }
  ]
  

3. ZeroShot Embeddings

• Creating a Jac Walker (get_embeddings)

  1. Create get_embeddings walker in the zs_classifier.jac

         walker test_get_embedding{
             can zs_classifier.get_embeddings;
             report zs_classifier.get_embeddings(texts="I am so glad you liked it!");
         }
     

     Parameter details

     • get_embeddings: can be used to get embeddings for text from zs model
       • Input:
         • text (str or List(str)): it can be string or list of string
       • Returns: list of embdedding of length 768 size

• Steps for running zs_classifier.jac program

  1. Build zs_classifier.jac by run cmd

     jac build zs_classifier.jac

  2. Activate sentinal by run cmd

     sentinel set -snt active:sentinel -mode ir zs_classifier.jir

  3. Calling walker get_embeddings with default parameter

     walker run get_embeddings
     

  Expected output

  [
  0.15722215175628662,
  -0.2446146011352539,
  0.7713489532470703,
  0.20605269074440002,
  ...
  ...
  ...
  ...
  -0.4917687475681305,
  0.17996184527873993,
  -0.39374977350234985,
  0.3616657555103302
  ]