Twigums/mal-anime-recommender

mal-anime-recommender

Steps to get this to work locally:

1. pip3 install -r requirements.txt

2. Get an api (I will not share my api token... :( )

   • run python3 mal_api.py get_api
   • follow instructions

3. Now that you have a client id and access token,

   • refreshing api: python3 mal_api.py refresh_api
   • you can choose what to do now

4. Retrieving anime ids:

   • python3 mal_api.py anime_ids param1 param2 param3
   • limit = "param1": how many top animes you want
   • init_offset = "param2": if you want to offset the ranking (ex: start at anime #50 -> init_offset = 50)
   • append_status = "param3": boolean (True/False) to determine if we want to append to an existing anime_ids file

5. Retrieving anime details:

   • requires anime ids
   • python3 mal_api.py anime_details param1
   • append_status = "param1": boolean (True/False) to determine if we want to append to an existing anime_ids file
   • follow instructions

6. Retrieving user information:

   • requires api
   • python3 mal_api.py user_info param1 param2
   • username = "param1": username to retrieve info from
   • append_status = "param2": boolean (True/False) to determine if we want to append to an existing user info file(s)

7. Retrieving videos:

   • requires video urls from anime details step
   • python3 video_parser.py get_videos

8. Trimming available videos:

   • requires videos
   • python3 mal_api.py trim_all param1 param2
   • start_percentage = "param1": to what percent to clip off
   • end_percentage = "param2": from what percent to clip off
   • if you only want 20-80% -> start_percentage, end_percentage = 20, 80

9. Generate random frames:

   • requires videos
   • python3 mal_api.py random_frames_all params1
   • frames = "params1": the number of frames to save per video

10. Removing violating frames

    • requires frames
    • python3 mal_api.py remove_bad_frames
    • function is really slow and does not remove every violating frame
    • highly recommended to go through yourself

11. Making ImageFolder

    • requires frames and classes (if move user info, 2nd line)
    • python3 image_to_tensor.py move_to_classes param1 param2
    • python3 image_to_tensor.py user_move_to_classes param1 param2 param3 param4
    • path_to_make = "param1": destination of where to make the ImageFolder
    • path_to_folder = "param2": where the current folder with frames is
    • path_user_labels = "param3": where the text file for user scores is
    • path_user_animes = "param4": where the text file for user anime is

12. (Optional) Make noisy images

    • requires frames
    • python3 image_to_tensor.py add_noise param1 param2 param3
    • path_to_images = "param1": where the current folder with frames is
    • path_to_noised_images = "param2": destination of where to make the new frames
    • std = "param3": standard deviation of the normal Gaussian distribution to specify
    • to replicate my set, std was 0.8 and 3.

13. Running general model

    • requires ImageFolder dataset and CUDA
    • python3 CNN.py
    • remember to specify filepaths
    • path_to_data is where you have the training dataset
    • path_to_test is where you have the testing dataset
    • path_to_model is where you save/load the model
    • additional params: need_save and noRound can be specified (default True/False respectively to save model and round anime scores)

14. Running user model

    • requires ImageFolder dataset, CUDA, saved model
    • python3 test_model.py
    • remember to specify filepaths
    • path_to_user_data is where you have the user dataset
    • path_to_recommend_data is where you have the small ImageFolder dataset
    • path_to_model is where you saved the general model
    • path_to_save_model is where you want to save/load the user model
    • additional params: need_save and noRound can be specified (default True/False respectively to save model and round anime scores)
    • if you want to have the model recommend, specify needs_save as False