Text experiments

[WillemSpek]

It's been a while since my last PR, but I haven't been idle. Just took me a while to get the text metrics to work. I have attempted to organize the grid search using XAI metrics. In general this PR contains the following contributions:

1. Script to test against multiple hyperparameter configurations for textual data.
2. Notebooks used to sample the original datasets according to model scores.
3. The sampled datasets themselves.
4. Improvement over the single deletion metric to test against a set of salient maps.
5. Addition of sensitivity metric for textual data.
6. A synonym replacer that attempts to perturb sentences by replacing words with their synonyms.
7. Notebooks used to analyze the collected data and produce visualizations for them.

From a software engineering perspective, there is definitely still quite a bit to be improved here as the code is now a bit bloaty and ill-organized. Nevertheless, I will try to get this work as organized as possible whilst still running the code needed to obtain my results.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[elboyran]

I've left few comments here and there.
Looks good to me in general (a lot of work), I admit I did not run the code.