README.md

FFM on the Criteo data

Try to replicate the result of Kaggle Display Advertising Challenge by using the following implementations of field-aware factorization machines (FFMs):

• LIBFFM
• Hivemall

Data

Download and convert the full dataset into CSV format:

./data.sh
ln -s train.csv tr.csv
ln -s test.csv te.csv

Or, use tiny data:

ln -s train.tiny.csv tr.csv
ln -s test.tiny.csv te.csv

Usage

Build LIBFFM and preprocess data as the winning solution did:

make

Move to libffm/ or hivemall/ and follow the instructions.

References

• guestwalk/kaggle-2014-criteo
• chenhuang-learn/ffm
• https://gist.github.com/myui/aaeef548a17eb90c4e88f824c3ca1bcd