Spaces:
Sleeping
Sleeping
File size: 2,567 Bytes
41b0868 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 |
import pandas as pd
import torch
import faiss
import numpy as np
from numpy import dot
from numpy.linalg import norm
def table_maker(
df: pd.DataFrame,
country: list = [],
min_year: int = 1999,
max_year: int = None,
tagger=set(),
rating: bool = True,
):
x = df.copy()
# фильтр по рейтингк
if rating:
rat_con = ~(x["rating"].isna())
else:
rat_con = ~(x["url"].isna())
# фильтр по стране
if country == []:
con_con = ~(x["url"].isna())
else:
con_con = x["county"].isin(country)
# фильтр по тегам
if tagger == set():
tagger_con = ~(x["url"].isna())
else:
tagger_con = x["tags"].ge(tagger)
# Условие для фильтрации по минимальному году
year_cond = x["year"] >= min_year
# Добавляем условие для фильтрации по максимальному году, если оно задано
if max_year is not None:
year_cond &= x["year"] <= max_year
condi = rat_con & con_con & tagger_con & year_cond
return x.loc[condi]
class RecSys:
def __init__(self, df: pd.DataFrame, input_, model):
self.df = df
self.input_ = input_
self.model = model
with torch.no_grad():
self.emb = model.encode(self.input_)
def __call__(self):
def compute(a):
return dot(a, self.emb) / (norm(a) * norm(self.emb))
res = self.df.copy()
res["compute"] = res["vec"].map(compute)
res["compute2"] = res["vec2"].map(compute)
self.df["score"] = res["compute"] * 0.8 + res["compute2"] * 0.2
return self.df.sort_values("score", ascending=False)
class FAISS_inference:
def __init__(self, df, emb, k=5):
self.df = df
self.emb = emb.reshape(1, -1)
self.k = k
vec = df["vec"].to_numpy()
self.d = vec[0].shape[0]
for i, e in enumerate(vec):
if i == 0:
vex = e.T
else:
temp = e.T
vex = np.append(vex, temp)
self.vex = np.reshape(vex, (-1, 384))
# self.index = faiss.IndexFlatIP(self.d)
# self.index = faiss.IndexFlatL2(self.d)
self.index = faiss.IndexFlat(self.d)
self.index.add(self.vex)
def __call__(self):
d, i = self.index.search(self.emb, self.k)
faiss_table = self.df.iloc[i[0]]
faiss_table.loc[:, "score"] = d[0]
return faiss_table
|