This is our code for the Project 1 of Machine Learning Fall 2021

• Team Name: TP-Link
• Team Member:

1. Silin Gao ([email protected])
2. Shaobo Cui ([email protected])
3. Dongge Wang ([email protected])

• Python
• Numpy
• Matplotlib

• implementations.py: implementations of baseline models and our improved model.
• processor.py: data processor, including data imputation, normalization, outlier filtering, feature augmentation and selection.
• run_baselines.py: script for training and evaluating baseline models, including cross validation and test set prediction.
• run.py: script for training and evaluating our improved model, including cross validation and test set prediction.
• toolkits.py: toolkits including data loader, batch generator, metrics computer, file writer, etc.
• plot_weights.py: script for plotting output weights of features in reg_logistic_regression (outlier factor = 10, polynomial factor = 4), used for feature selection.

Place the original training and test sets (train.csv and test.csv) in the root directory of our project.

python processor.py

Outputs:

• filter_factor_${outlier_factor}_y.csv: training labels after outlier filtering.
• filter_factor_${outlier_factor}_tx.csv: training features after outlier filtering.
• filter_factor_${outlier_factor}_poly_${polynomial_factor}_tx.csv: training features after outlier filtering and feature augmentation.
• select_feature_top20_filter_factor_${outlier_factor}_poly_${polynomial_factor}_tx.csv: training features after outlier filtering, feature augmentation and feature selection.
• test_id.csv: sample ids in test set.
• test_tx.csv: original test features.
• test_poly_${polynomial_factor}_tx.csv: test features after feature augmentation.

python run_baselines.py

Note: Baseline running directly loads and pre-processes the original data (train.csv and test.csv), which is independent from the data processing. Outputs:

• results_${model_name}_k_${cross_validation_folds}.csv: cross validation results.
• results_${model_name}.csv: final prediction results on test set.

python run.py

Outputs:

• results_reg_logistic_dynamic_k_${cross_validation_folds}_poly_${polynomial_factor}.csv: cross validation results.
• results_reg_logistic_dynamic_poly_${polynomial_factor}.csv: final prediction results on test set (best submission).

We include the test set prediction results of all baseline models and our improved model under the folder "predictions".