def get_similar_user(user_id, nb_user=6):
user_id = int(user_id)
factors = load_pickle('save/user_factors.pkl')
distances = torch.nn.CosineSimilarity(dim=1)(factors,
factors[user_id][None])
idx = distances.argsort(descending=True)[1:nb_user + 1].numpy()
return list(idx)
def get_similar_book(title,
series=None,
authors=None,
volume_number=None,
nb_books=15):
book_id = get_book_id(title, series, authors, volume_number)
factors = load_pickle('save/book_factors.pkl')
distances = torch.nn.CosineSimilarity(dim=1)(factors,
factors[book_id][None])
idx = distances.argsort(descending=True)[1:nb_books + 1]
idx = idx[1:nb_books + 1]
titles = pd.read_csv('save/titles_by_book_id.csv', header=None)[0].values
return list(titles[idx])
def tabular_predict_from_nn(tab_fn, weights_fn, xs=None):
#learner = load_learner(model_fn)
to_nn = fasttab.load_pickle(tab_fn)
dls = to_nn.dataloaders(1024)
learn = fasttab.tabular_learner(
dls, metrics=fasttab.accuracy
) #BrierScore doesn't seem to work with lr_find()
learn.load(weights_fn)
if not isinstance(xs, pd.DataFrame):
return learn
dl = learn.dls.test_dl(xs, bs=64) # apply transforms
preds, _ = learn.get_preds(dl=dl) # get prediction
return preds
def get_rating_user_book(user_id,
title,
series=None,
authors=None,
volume_number=None):
user_id = int(user_id)
book_id = get_book_id(title, series, authors, volume_number)
model = load_pickle('save/nn_collab.pkl')
data = pd.read_csv('save/book_nn_data.csv')
data.set_index('book_id', drop=False, inplace=True)
row = data.loc[[book_id]].copy()
row['user_id'] = user_id
dl = model.dls.test_dl(row)
rating, _ = model.get_preds(dl=dl)
return rating.item()
def get_book_avg(author,
pub_year,
language_code,
genres="",
series=None,
volume_number=None):
genres = genres.split(',')
model = load_pickle('save/avg_rating_nn.pkl')
all_genres = pd.read_csv('save/genres.csv', header=None)[0].values
row = {genre: False for genre in all_genres}
for genre in genres:
row[genre] = True
row["authors"] = author
row["original_publication_year"] = int(pub_year)
row["language_code"] = language_code
row["series"] = series
row["volume_number"] = volume_number
*_, rating = model.predict(pd.Series(row))
return rating.item()