eals_experience.py

from amplify import IsingPoly, IsingMatrix, gen_symbols, sum_poly, BinaryPoly
from amplify import Solver, decode_solution
from amplify.client import FixstarsClient
import numpy as np
import pandas as pd
import scipy.sparse as sps
from itertools import combinations
from tqdm import tqdm
from eals import ElementwiseAlternatingLeastSquares
from sklearn.metrics import roc_auc_score, accuracy_score
import matplotlib.pyplot as plt

df = pd.read_csv("sample.csv").iloc[:, :]
# 上から1000行使う
train_df = df.iloc[:1000, :]
# 縦持ち
long_df = pd.melt(train_df, id_vars=["id"], value_vars=train_df.iloc[:, 1:].columns)
# 数値に変換
long_df["id"] = long_df["id"].apply(lambda x: x[2:]).astype(int)
long_df["variable"] = long_df["variable"].apply(lambda x: x[1:]).astype(int)

matrix = sps.csr_matrix(
    (long_df["value"], (long_df["id"], long_df["variable"])),
)
model = ElementwiseAlternatingLeastSquares()
model.fit(matrix, show_loss=True)

result = np.zeros(
    # train_dfはid列があるので-1する
    (1000, train_df.shape[1] - 1)
)

# predict
for i in range(1000):
    user_vector = model.user_factors[i]
    pred_ratings = model.item_factors @ user_vector

    result[i, :] = pred_ratings

score = pd.DataFrame(columns=train_df.iloc[:, 1:].columns)

for i in range(train_df.shape[1] - 1):
    # train_dfはid列があるのでi + 1する
    true = train_df.iloc[:, i + 1]
    pred = result[:, i]
    print(f"\nx{i}")
    try:
        # 正例が無いとエラーになるから飛ばす
        print(roc_auc_score(true, pred))
        score.loc["train_auc", f"x{i}"] = roc_auc_score(true, pred)
    except:
        continue

    score.to_csv("eals_result.csv")