iampara0x / yuno Goto Github PK

It would be awesome to retrain Yuno every once in a while, as new anime and recommendations appear. Even once a year would be good to have.

good job!
but result seem not so good.

I know many Recommendation Systems based on movie entity graph (meta data) and user comment (based on NLP)

and I think you can give it a try on synopsis (japanese article) reference from this

and I just discovered tmdb can be a good source , seem English synopsis edit almost on every anime entity, and its parallel multilingual text
some well-known example:
Story synopsis - API
epsiode synopsis or interface language mode=ja-JP - API

if Interested on multilingual sbert multilingual model

I didn't found who build a Parallel text dataset or dump with translations dev reference
you can use discover API to get anime tv/moive id

Cannot run program "D:\PythonProject\Yuno\Pipfile" (in directory "D:\PythonProject\Yuno"): CreateProcess error=193, %1 It isn't available Win32 Application。

I get this error everytime

<ipython-input-2-4604c83e7d7e> in <module>
      1 import dill as pickle
----> 2 from yuno.search import pipelines,base,config,utils
      3 import torch
      4 import torch.nn.functional as F
      5 import faiss

ModuleNotFoundError: No module named 'yuno'

Hi, I can see that you did a great job on Yuno! As I'm on my way to figuring out how this model works and its preprocessing steps, I engaged some problems with the preprocessing part.

Currently I'm looking at the filter.py in the preprocessing folder. The class FilterText is initialized with AnimeInfo class.

class AnimeInfo(NamedTuple):

  uid: int

  names: List[str]

  characters: List[Character]

As I see, it contains anime uid(it should match MAL's uid), anime title and character info. But I couldn't find the character data in the Kaggle dataset or anime character scraper for MAL or anything. So that raised me some questions.

1. How did you attain the character data ? Is there any notable preprocessing needed to be done after fetching the data ?

2. What is included in the info for each anime character ? (their name, gender, age, ...etc)

By the way I think it'd be nice if you upload another notebook for the preprocessing part. You may be busy, so I don't mind if you don't. Thanks in advance!

FileNotFoundError: [Errno 2] No such file or directory: 'search_base.pkl'

And

NameError: name 'pipeline' is not defined

Hi, I find your work very inspiring and the program code very complete. I would like to learn your special way of training. It looks like you used an unsupervised approach (maybe called Pair-based metric learning) to train the model to learn semantics. I am curious how pos, neg and anchor are selected. I'm still reading the code. I haven't fully understood it yet.

According to my understanding, the comments of the same anime are set as anchor and pos, while the comments of any other randomly selected anime are defined as neg