flamingofugang / causality

Kaggle's Causality Challenge Solution for team FirfiD

Cause Effect Pairs Challenge FirfiD Submission

Pre-requisites: You need the following installed: python 2.7.1 python sklearn version 0.13.1 python numpy version 1.7.1 python joblib python pandas version >=0.11

Matlab Preferably Debian Based Linux Installation

Kaggle Causality Challenge framework. Mostly based on kaggle's python code code for the challenge.

To train: A. Configure

1. Put your training data in the following files (or modify file names accordingly):

"train_pairs_path": "./Competition/CEdata_final_train_pairs.csv" "train_info_path": "./Competition/CEdata_final_train_publicinfo.csv" "train_target_path": "./Competition/CEdata_final_train_target.csv"

B. Extracting Features

1. Modify SETTINGS.json "feature_extraction_threads" to the number of threads your machine can handle.
2. Run "python fe.py"
3. Add Matlab features by running "./extract_matlab_valid.sh"
4. Merge the futures by running "python process_matlab.py -t valid"

C. Train:

1. Run "python train.py"

To predict:

A. Clean-up

1. Replace ./Competition/CEfinal_valid*.csv with the respective files you are interested in extracting features from. By default this is set to a minimal subset of valid features.
2. Run "./clean.sh"

B. Extracting Features

1. Modify SETTINGS.json "feature_extraction_threads" to the number of threads your machine can handle.
2. Run "python fe.py"
3. Add Matlab features by running "./extract_matlab_valid.sh"
4. Merge the futures by running "python process_matlab.py -t valid"

C. Generating results

5. Run "python predict.py". The results file should be "./Submisions/firfi-tree-trees.csv".