A Full Scale Movie Recommendation Engine In Python And Numpy
(Note: This project makes use of Git LFS for managing large datasets. So viewing the files in the browser wouldn't be possible.)
For a one stop solution, I always use conda. First install conda and create you own environment as:
conda create --name myenv
where myenv is the environment name (which can be changed at will). Then activate the environment by:
source activate myenv
and then install the dependencies using the following command:
conda install -c conda-forge matplotlib numpy pandas theano
You need to clean the given dataset. For MovieLens 10M dataset, you can just run the following command in the src folder:
python3 preprocess.py
which would create a dense dataset file named ratings_cleaned.csv, and the training and test sets as train.csv and test.csv in the MovieLens 10M Dataset folder.
You can then execute the project using the following command:
python3 main.py
All the plots would be saved under plots folder. Additionally, you can change the parameter and the test conditions for various tests in main.py file to suit your needs. Be cautioned, many of the test conditions implemented would take days to finish on an average machine. Most of the plots shown were computed using Google Cloud Platform.
The official Netflix Prize dataset hasn't been uploaded for the size related reasons, but can be downloaded from here. It should be noted that the dataset should work out of the box, provided the above mentioned instructions are followed. Lastly, for testing on any new dataset, please ensure that the dataset is structured in a similar manner as the ratings.csv file provided.