Open the Technical_Report.ipynb file to view the full report.
- The scope of explainability require different approaches. What works locally may not work globally.
- Global explainations are better done with tokens clustered into topics.
- Local explainations are better without decomposition for simpler models.
- The easiest local explanation is the addition of token weights.
- The Pipeline class can handle vectorization and decomposition in succession.
In partial fulfillment of the requirements of AIM MSDS 2022 Machine Learning 2 under Prof. Chris Monterola.