/kaggle-facebook-bot

Solution to the Kaggle Facebook Recruiting IV contest

Primary LanguageJupyter NotebookMIT LicenseMIT

kaggle-facebook-bot

This repository contains a Jupyter notebook that outlines my approach for the Kaggle Facebook Recruiting IV contest.

The .csv-files of predictions can be generated as follows:

  1. Download and extract the data from https://www.kaggle.com/c/facebook-recruiting-iv-human-or-bot/data. Place the files train.csv, test.csv and bids.csv into the data-directory of this repository. Place the file sampleSubmission.csv into the submissions-directory of this repository.

  2. Run the facebook_notebook.ipynb-notebook. This should generate the submission filefacebook_submission.csv into the submissions-directory.

Libraries:

The basic scientific Python libraries + XGBoost

Running time/Hardware:

Runs in about 15 minutes on a fairly high-powered desktop (i7-4790) with 16 gb of RAM. Can clog up the ram on smaller machines.

Update Feb 21st 2016

  • Added a Model interpretation-section to the notebook
  • Added a hyperopt_xgb.py-script that shows how hyperparameters can be optimized using a grid search.
    • The script generates a file hyperopt_xgb.csv in the root of the repository which displays a selection of hyperparameters and the corresponding cross-validated AUC-score.
    • Running the script requires two additional dependencies: the hyperopt- and pymongo-libraries.