fdoperezi/kaggle-coupon-purchase-prediction-1

map@10 0.0072 Kaggle-Recruit Coupon Purchase Prediction Challenge

Title	Author	First Created
Kaggle Coupon Purchase Prediction Challenge	Aakansh Gupta	October 5, 2015

This repo contains code for the Recruit Ponpare Coupon Prediction Challenge. It gives map@10 score 0.0072 on Private Leaderboard.

This was executed and tested on r3.8xlarge ami on AWS. Execution requires R 3.2.2 along with the package xgboost and others as mentioned in RunME.R.

To run do the following:

1. Download the data
2. Download the repo at the same location.
3. Create a new folder 'raw_data' and unzip the data files in the folder.
4. Install packages as required in RunMe.R
5. Modify paths and run RunsMe.R
6. A submission folder is then generated containing the submissions.

NOTE on using MyMediaLite for Matrix Factorization:

The binaries of the required version of MyMediaLite are already cloned inside the code folder in this repo. Just fulfill the requirements for running MyMediaLite:

• Mono 2.8.x or another recent .NET runtime (Mono 2.10.x recommended)

Details about the implementation can be found in this blog post: http://datascience.blog.uhuru.co.jp/machine-learning/predicting-coupon-purchases-on-ponpare/