CriteoLab Click-Through-Rate

Course: w261 Final Project (Criteo Click Through Rate Challenge)

Team Number:19

Team Members: Steve Dille, Naga Akkineni, Joanna Yu, Pauline Wang

Fall 2019: section 1, 3, 4

Our final submission comprised of the following components:

• Jypter notebook - main Jupyter notebook of our project.
• .py files - Python files used for GCP submission to run the models in the cloud.

1. 'steve_ctr_DT_full.py' - GCP job for Decision Tree base model (Model I) on the full training set.
2. 'final_proj_GCP_RF_SSI2.py' - GCP job for Random Forest base model (Model II) on the toy dataset.
3. 'steve_ctr_RF_full.py' - GCP job for Random Forest base model (Model II) on the full training set.
4. 'steve_ctr_ ' - GCP job for Random Forest model with preprocessing, meaning scaled integer features and string indexed categoricl features (Model III) on the full training set.
5. 'steve_crt_RF_full.py' - GCP job for Random Forest model with preprocessing listed in Model III plus Gradient Boosting (Model IV) on the full training set.

• Pickle files - Pickle files for the Pandas tables that we created for pretty printing:

1. 'summary.pkl' stores the summary statistics table for the training dataset.
2. 'correlation.pkl' stores all the pairwise correlation values for all 39 features and the label for the training dataset.
3. 'correlation_subset.pkl' takes the table from 'correlation.pkl' and filter out entries with correlation values > 0.5.
4. 'DT_base_PD.pkl' stores the summary metrics table for the Decision Tree base model with varied MaxDepth to compare model performance.
5. 'toyDF_PD.pkl' stores the Pandas dataframe for the toy dataset used for EDA. The toy dataset is 1.5% of the training set.