Awesome

Efficient Collaborative Filtering with LSH

Quick link to the project report.

Requirements

Python 2.7.x or 3.3.x
numpy + scipy modules
Download and extract data.zip (720MB) or data.7z (460MB) into the root directory of the project

Instructions

Unzip the contents of compressed file data.(zip|7z) into the root directory. There should now be files movie_titles.txt, probe.txt and folders training_set, training_med in the root directory.
cd to src/
Run python gen.py with appropriate command line arguments to generate disjoint training and test sets. python gen.py -h gives further information.
Run python main.py with appropriate command line arguments (-h flag for help). This will create the nearest neighbour data structure using the training set you generated in the previous command, then it will attempt to predict ratings for test set users and calculate the error.
The last line printed by the script is the RMSE on the probe set.

The k,l parameters described in the report can be configured as constants in src/config.py.

Sample Usage

antares: src\ $ python gen.py 300
successfully generated datasets.
antares: src\ $ python main.py 300
index progress: 0.000%
index progress: 0.917%
index progress: 1.835%
...
index progress: 97.248%
index progress: 98.165%
index progress: 99.083%
indexing and setup complete
evaluation progress: 0.000%
evaluation progress: 1.010%
evaluation progress: 2.020%
...
evaluation progress: 96.970%
evaluation progress: 97.980%
evaluation progress: 98.990%
RMSE: 1.05259425337