Movie Recommendation Engine

Using SVD algorithm for collaborative filtering.

• First train a model by using available ratings
• Use that trained model to predict missing rating in users vs items/movies matrix
• With all pridicted rating now users vs items/movies matrix become trained users vs items/movies matrix and we save both in formof .pkl file.
• Later we use any of users vs items/movies matrix or trained users vs items/movies matrix by TRAINED argument,for further processing.

First you need to train the model

python train.py

 --data_file       : Input user-movie-rating information file (default: './ratings.dat')
 --batch_size      : Batch Size (default: 100)
 --dims"           : Dimensions of SVD (default: 15)
 --max_epochs      : Dimensions of SVD (default: 25)
 --checkpoint_dir  : Checkpoint directory from training run (default: '/save/')
 --val             : True if Folders with files and False if single file
 --is_gpu          : Want to train model at GPU (default=True)

Misc Parameters:

--allow_soft_placement    :Allow device soft device placement
--log_device_placement   :Log placement of ops on devices

• After training finish, we need to save trained model as well as trained user vs item matrix for later use for recommendation.

Run.py file predict the rating for a given user and movie pair

python run.py

  --user            : User (default: 1696)")
  --item            :Movie (default: 3113)")
  --checkpoint_dir  : Checkpoint directory from training run (default: '/save/')
  --is_gpu          : Want to train model at GPU (default=True)

Misc Parameters :

--allow_soft_placement    :Allow device soft device placement
--log_device_placement   :Log placement of ops on devices

For find the K-mean clusters

python kmean.py

  --data_file       : Input user-movie-rating information file (default: './ratings.dat')
  --K               : Number of clusters (default=4)
  --MAX_ITERS       : Maximum number of iterations (default=1000)
  --TRAINED         : Use TRAINED user vs item matrix (default=False)

output of kmean.py saved in clusters.csv file

Note: Rest How to use this please go through poc.ipynb file