data/func.py

import pandas as pd
import numpy as np
import torch
from transformers import BertModel, BertTokenizer
from sklearn.metrics.pairwise import cosine_similarity


tokenizer = BertTokenizer.from_pretrained("DeepPavlov/rubert-base-cased-sentence")
model = BertModel.from_pretrained("DeepPavlov/rubert-base-cased-sentence")
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")


def filter_by_ganre(df: pd.DataFrame, ganre_list: list):
    filtered_df = df[df['ganres'].apply(lambda x: any(g in ganre_list for g in(x)))]
    filt_ind = filtered_df.index.to_list()
    return filt_ind

# def mean_pooling(model_output, attention_mask):
#     token_embeddings = model_output['last_hidden_state']
#     input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
#     sum_embeddings = torch.sum(token_embeddings * input_mask_expanded, 1)
#     sum_mask = torch.clamp(input_mask_expanded.sum(1), min=1e-9)
#     return sum_embeddings / sum_mask

# def recommendation(filt_ind: list, embeddings: np.array, user_text: str, n=10):
#     token_user_text = tokenizer(user_text, return_tensors='pt', padding='max_length', truncation=True, max_length=512)
#     user_embeddings = torch.Tensor().to(device)
#     model.to(device)
#     model.eval()
#     with torch.no_grad():
#         batch = {k: v.to(device) for k, v in token_user_text.items()}
#         outputs = model(**batch)
#         user_embeddings = torch.cat([user_embeddings, mean_pooling(outputs, batch['attention_mask'])])
#         user_embeddings = user_embeddings.cpu().numpy()
#     cosine_similarities = cosine_similarity(embeddings[filt_ind], user_embeddings.reshape(1, -1))
#     df_res = pd.DataFrame(cosine_similarities.ravel(), columns=['cos_sim']).sort_values('cos_sim', ascending=False)
#     dict_topn = df_res.iloc[:n, :].cos_sim.to_dict()
#     return dict_topn


def embed_user(filt_ind: list, embeddings:np.array, user_text: str, n=10):
    tokens = tokenizer(user_text, return_tensors="pt", padding=True, truncation=True).to(device)
    model.to(device)
    model.eval()
    with torch.no_grad():
        outputs = model(**tokens)
        user_embedding = outputs.last_hidden_state.mean(dim=1).cpu().numpy().reshape(1, -1)
    return user_embedding

    # cosine_similarities = cosine_similarity(embeddings[filt_ind], user_embedding.reshape(1, -1))
    # df_res = pd.DataFrame(cosine_similarities.ravel(), columns=['cos_sim']).sort_values('cos_sim', ascending=False)
    # dict_topn = df_res.iloc[:n, :].cos_sim.to_dict()
    # return dict_topn