James Wexler

Thanks for the report. I'm wondering if when you do "pip list" if pyparsing version is not 2.4.7 but 3.x. If you are really using 2.4.7 then I wouldn't expect...

What is the notebook environment you are seeing this failure in? Does restarting the kernel help?

Are you able to test with python 3.6 or 3.7? I haven't tried WIT in py 3.9 yet.

I believe we haven't tested WIT in JupyterLab 3.x and it may not work. Have you attempted JupyterLab 2.x and if so, does that work for you? Please see https://github.com/PAIR-code/what-if-tool/issues/164...

WIT should be able to handle all those input types in a single datapoint just fine. Your code might look like some combination of https://colab.sandbox.google.com/github/PAIR-code/what-if-tool/blob/master/WIT_Smile_Detector.ipynb (images), https://colab.sandbox.google.com/github/pair-code/what-if-tool/blob/master/WIT_Toxicity_Text_Model_Comparison.ipynb (text), and https://colab.sandbox.google.com/github/pair-code/what-if-tool/blob/master/WIT_COMPAS_with_SHAP.ipynb...

Thanks for reaching out @JonnyCBB . I've never tried it with a TFRanking model, but if the predictions can be represented as either a classification or regression output then there's...

Oh well. If you could send over a file with just a few fake examples in the same format I can investigate. No hurry though.

In that case, you may need to create a separate version of those records, but with readable-string versions of those values through some encoding. Then in your custom prediction function,...

Didn't plan on any support since this is the first time its come up, but please open an issue in this repo and we can track it there.

Unfortunately, WIT only works for simple classification or regression tasks. It doesn't have support for handing bounding boxes as outputs. The Language Interpretability Tool (https://github.com/PAIR-code/lit/, https://pair-code.github.io/lit/) is a successor of...