AxisError: axis 1 is out of bounds for array of dimension 1 about multimodal-toolkit

[akashsara]

Hey @apethree, you've hit the solution there. Due to a change in how EvalPrediction works, the calc_classification_metrics is bugged. The variable with the 768 refers to the embeddings before the final layer. Just use p.predictions[0] instead and you should be good.

[apethree]

I have made progress. The shapes of the arrays are:

p.predictions of the type:

p.predictions[0] -->. Array of size (1200,2)
p.predictions[1] --> List [0], [1]
Where [0] ---> (1200,768)
Where [1] ---> (1200,2)

What is the 768 wide row? Not sure how to use that in prediction and F1 score.

[apethree]

This the probably the correct way to calculate argmax():

  pred_labels = np.argmax(p.predictions[0], axis=1)
  pred_scores = softmax(p.predictions[0], axis=1)[:, 1]

Instead of

  pred_labels = np.argmax(p.predictions, axis=1)
  pred_scores = softmax(p.predictions, axis=1)[:, 1]