Scripts
testing_package.jl
Contains all the basic functionality of the package:
- Loading the dataset
- Loading the weights or training a new model
- Evaluating the model on the provided test set.
filter_movielens.jl
An ugly script to extract needed training and testing data from the movielens dataset. Leaving code as it is because it would always be different for different datasets.
Scaled the rankings from 1 to 5 stars to <0.1, 1> by MinMaxScale.
Splitted data to the train/test randomly.
Resulting dataset that is used in the NeuralCollaborativeFiltering training is of form:
| user | movie | score |
|---|---|---|
| 1 | 47 | 1.0 |
| 2 | 256 | 0.7 |
...
Also, I extracted the movie/genre pairs, which could potentially be used in the recommendation system that would consider those as additional features (see README).