Collaborative Filtering is a method used by recommender systems to make predictions about an interest of an specific user by collecting taste or preferences information from many other users. The technique of Collaborative Filtering has the underlying assumption that if a user A has the same taste or opinion on an issue as the person B, A is more likely to have B’s opinion on a different issue.
In this project I use a Restricted Boltzmann to predict whether a user would like a movie or not based on this user's taste and the taste of other users who watched and rated the same and similar movies.
- Python 3.6
- TensorFlow 1.5 or higher
- NumPy 1.11 or higher
The current version support only the MovieLens ml-1m.zip dataset obtained from https://grouplens.org/datasets/movielens/.
-
Download the ml-1m.zip dataset from https://grouplens.org/datasets/movielens/.
-
Save ratings.dat under DATA_DIR. Devide the ratings.dat into training and testing datasets train.dat and test.dat with the shell command:
python train_test_split.py DATA_DIR OUTPUT_DIR_TRAIN OUTPUT_DIR_TEST -
Use shell to make TF_Record files out of the both train.dat and test.dat files by executing the command:
python tf_record_writer.py OUTPUT_DIR_TRAIN OUTPUT_DIR_TEST OUTPUT_DIR_TF_TRAIN OUTPUT_DIR_TF_TEST -
Use shell to start the training by executing the command (optionally parse your hyperparameters):
python train.py \ --tf_records_train_path=OUTPUT_DIR_TF_TRAIN \ --tf_records_test_path=OUTPUT_DIR_TF_TEST \
During the training process we can examine the progress of the accuracy on training and test sets. The accuracy gives the ratio of correctly predicted binary movie ratings. It can be seen that after 16 epochs the model predicts 78,4% of the time correctly if a user would like a movie or not.
epoch_nr: 0, batch: 50/188, acc_train: 0.721, acc_test: 0.709
epoch_nr: 0, batch: 100/188, acc_train: 0.744, acc_test: 0.704
epoch_nr: 0, batch: 150/188, acc_train: 0.748, acc_test: 0.736
epoch_nr: 1, batch: 50/188, acc_train: 0.767, acc_test: 0.764
epoch_nr: 1, batch: 100/188, acc_train: 0.758, acc_test: 0.733
epoch_nr: 1, batch: 150/188, acc_train: 0.756, acc_test: 0.732
epoch_nr: 2, batch: 50/188, acc_train: 0.772, acc_test: 0.773
epoch_nr: 2, batch: 100/188, acc_train: 0.758, acc_test: 0.769
epoch_nr: 2, batch: 150/188, acc_train: 0.754, acc_test: 0.771
epoch_nr: 3, batch: 50/188, acc_train: 0.767, acc_test: 0.725
epoch_nr: 3, batch: 100/188, acc_train: 0.758, acc_test: 0.757
epoch_nr: 3, batch: 150/188, acc_train: 0.756, acc_test: 0.760
epoch_nr: 4, batch: 50/188, acc_train: 0.768, acc_test: 0.717
epoch_nr: 4, batch: 100/188, acc_train: 0.756, acc_test: 0.743
epoch_nr: 4, batch: 150/188, acc_train: 0.759, acc_test: 0.781
epoch_nr: 5, batch: 50/188, acc_train: 0.772, acc_test: 0.769
epoch_nr: 5, batch: 100/188, acc_train: 0.762, acc_test: 0.774
epoch_nr: 5, batch: 150/188, acc_train: 0.760, acc_test: 0.775
epoch_nr: 6, batch: 50/188, acc_train: 0.774, acc_test: 0.771
epoch_nr: 6, batch: 100/188, acc_train: 0.764, acc_test: 0.776
epoch_nr: 6, batch: 150/188, acc_train: 0.765, acc_test: 0.775
epoch_nr: 7, batch: 50/188, acc_train: 0.779, acc_test: 0.780
epoch_nr: 7, batch: 100/188, acc_train: 0.765, acc_test: 0.778
epoch_nr: 7, batch: 150/188, acc_train: 0.766, acc_test: 0.775
epoch_nr: 8, batch: 50/188, acc_train: 0.777, acc_test: 0.785
epoch_nr: 8, batch: 100/188, acc_train: 0.768, acc_test: 0.785
epoch_nr: 8, batch: 150/188, acc_train: 0.764, acc_test: 0.766
epoch_nr: 9, batch: 50/188, acc_train: 0.779, acc_test: 0.787
epoch_nr: 9, batch: 100/188, acc_train: 0.763, acc_test: 0.784
epoch_nr: 9, batch: 150/188, acc_train: 0.765, acc_test: 0.783
epoch_nr: 10, batch: 50/188, acc_train: 0.775, acc_test: 0.763
epoch_nr: 10, batch: 100/188, acc_train: 0.766, acc_test: 0.750
epoch_nr: 10, batch: 150/188, acc_train: 0.771, acc_test: 0.784
epoch_nr: 11, batch: 50/188, acc_train: 0.776, acc_test: 0.768
epoch_nr: 11, batch: 100/188, acc_train: 0.765, acc_test: 0.783
epoch_nr: 11, batch: 150/188, acc_train: 0.767, acc_test: 0.789
epoch_nr: 12, batch: 50/188, acc_train: 0.783, acc_test: 0.789
epoch_nr: 12, batch: 100/188, acc_train: 0.768, acc_test: 0.762
epoch_nr: 12, batch: 150/188, acc_train: 0.764, acc_test: 0.774
epoch_nr: 13, batch: 50/188, acc_train: 0.776, acc_test: 0.786
epoch_nr: 13, batch: 100/188, acc_train: 0.767, acc_test: 0.778
epoch_nr: 13, batch: 150/188, acc_train: 0.764, acc_test: 0.785
epoch_nr: 14, batch: 50/188, acc_train: 0.777, acc_test: 0.786
epoch_nr: 14, batch: 100/188, acc_train: 0.764, acc_test: 0.765
epoch_nr: 14, batch: 150/188, acc_train: 0.767, acc_test: 0.767
epoch_nr: 15, batch: 50/188, acc_train: 0.778, acc_test: 0.786
epoch_nr: 15, batch: 100/188, acc_train: 0.769, acc_test: 0.783
epoch_nr: 15, batch: 150/188, acc_train: 0.770, acc_test: 0.783
epoch_nr: 16, batch: 50/188, acc_train: 0.783, acc_test: 0.790
epoch_nr: 16, batch: 100/188, acc_train: 0.766, acc_test: 0.787
epoch_nr: 16, batch: 150/188, acc_train: 0.767, acc_test: 0.786