kaggle-otto-challenge

Challenge Kaggle of Otto Products classification, for course TC1 of master AIC at Université Paris-Sud.

Autors : M. Bauw, N. Cadart, B. Sarthou

Date : November 2018

Use instructions

Copy kaggle otto challenge datasets into the data/ directory. Files can be downloaded here.
Run python3 final_training.py to train selected models, print results and save them to .csv file respecting kaggle submission format.

The execution may last about 10 minutes.

The final results should reach a log-loss score around 0.457.

data_exploration.py : compute and display information about the dataset : PCA, T-SNE, features importance, correlation matrix, sparsity...
data_preprocessing.py : script ot test several ideas on data classification, such as features normalization, features selection, dimension reduction or distribution probabilities calibration.
toolbox.py : useful tools functions for data parsing
gridsearch_training.py: run a grid search on several classification models with different hyper-parameters and save results to .csv files in directory gridsearch_results/.
final_training.py : train the selected models, find ensemble weights, print validation set results, run final predictions on kaggle test sets and save them in .csv file.
data/ : directory where datasets, results or images are saved.
gridsearch_results/ : directory where grid search results are savec when running gridsearch_training file.