DALEXtra with textual data

[giuliaass]

Hi, is it possible to explain a keras model (CNN) that has textual data as input with DALEX? I've only seen examples with tabular data. I preprocess the textual data with word embeddings and use them as a input for my CNN.

Thank you in advance!

[maksymiuks]

Inputs for methods served by DALEX and DALEXtra are assumed to be tabular. Could you please elaborate more about the format of your embedded inputs? I'll try to find sth that can help you

[giuliaass]

Hi,

To create my code I followed this example:

https://github.com/jjallaire/deep-learning-with-r-notebooks/blob/master/notebooks/6.1-using-word-embeddings.Rmd

First I have a raw text, then I tokenize the data. I'm also using pretrained embeddings.

Thank you! :-)

[hbaniecki]

For explanations of neural networks, you can check Captum for Python, and a new package innsight for R.