WIT missing functionality with TFX exports about what-if-tool

[jameswex]

Thanks for reaching out. Updates to get WIT in that notebook to work:

1. Change prediction function as it takes in an array of examples, not a single example. Also extract the list of class scores from the returned dict and return it.
   def predict(examples):
   preds = model.signatures['serving_default'](examples=tf.constant([example.SerializeToString() for example in examples]))
   return preds['outputs'].numpy()

2. Set up WitConfigBuilder to indicate this is a multi-class model:
   config_builder = WitConfigBuilder(eval_examples).set_custom_predict_fn(predict).set_multi_class(True)

Let me know if this works as expected for you.

[catherinenelson1]

Thank you! That works as expected.