Trying a simple ML experiment to check if the performance improves using Intel powered XGBoost
- Reads input csv file containing a text and its label
- Uses XGBoost to implement Single Label Text Classification Model
- Preprocessing steps are eliminated in this test to focus on the crux of the comparison
Steps to run the Intel extended XGBoost:
- Start running aikit_demo.ipynb notebook
- It sets up aikit-modin library and the Intel extension for XGBoost
- As the last step, it runs the aikit_xgboost_incidents notebook that actually contains the XGBoost model
- Make sure the incident_classification_postprocess.csv is present in the same path as the aikit_demo notebook
Step to compare that against the normal XGBoost:
- Open another instance (connecting to a different runtime session that does not have the aikit installed)
- Clone aikit_xgboost_incident.ipynb and the incident_classification_postprocess.csv files to it
- Run the notebook
- Note the time elapsed in "model.fit" step in both cases and compare results!