[논문구현]Factorization Machines(2010)

September 15, 2021

Factorization Machine

[논문 리뷰]Factorization Machines(2010)
1. Paper
논문
Higher-Order Factorization Machines
1. Reference: Open-source
libfm, libfm-github
PyTorchFM
fastFM
xlearn : 빠른 속도가 장점

이 아래는 FM 이해를 위해 FM 뼈대를 직접 구현해본거

Configuration

import os
import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
from sklearn.model_selection import train_test_split
import scipy
import math
from tqdm import tqdm
import warnings
warnings.filterwarnings("ignore")

from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive

data_path = "/content/drive/MyDrive/capstone/data/kmrd"
%cd $data_path

if not os.path.exists(data_path):
  !git clone https://github.com/lovit/kmrd
  !python setup.py install
else:
  print("data and path already exists!")

/content/drive/MyDrive/capstone/data/kmrd
data and path already exists!

path = data_path + '/kmr_dataset/datafile/kmrd-small'

Data Loader

KMRD

rates.csv

df = pd.read_csv(os.path.join(path,'rates.csv'))
train_df, val_df = train_test_split(df, test_size=0.2, random_state=1234, shuffle=True)

movie dataframe

# Load all related dataframe
movies_df = pd.read_csv(os.path.join(path, 'movies.txt'), sep='\t', encoding='utf-8')
movies_df = movies_df.set_index('movie')

castings_df = pd.read_csv(os.path.join(path, 'castings.csv'), encoding='utf-8')
countries_df = pd.read_csv(os.path.join(path, 'countries.csv'), encoding='utf-8')
genres_df = pd.read_csv(os.path.join(path, 'genres.csv'), encoding='utf-8')

# Get genre information
genres = [(list(set(x['movie'].values))[0], '/'.join(x['genre'].values)) for index, x in genres_df.groupby('movie')]
combined_genres_df = pd.DataFrame(data=genres, columns=['movie', 'genres'])
combined_genres_df = combined_genres_df.set_index('movie')

# Get castings information
castings = [(list(set(x['movie'].values))[0], x['people'].values) for index, x in castings_df.groupby('movie')]
combined_castings_df = pd.DataFrame(data=castings, columns=['movie','people'])
combined_castings_df = combined_castings_df.set_index('movie')

# Get countries for movie information
countries = [(list(set(x['movie'].values))[0], ','.join(x['country'].values)) for index, x in countries_df.groupby('movie')]
combined_countries_df = pd.DataFrame(data=countries, columns=['movie', 'country'])
combined_countries_df = combined_countries_df.set_index('movie')

movies_df = pd.concat([movies_df, combined_genres_df, combined_castings_df, combined_countries_df], axis=1)

print(movies_df.shape)
print(movies_df.head())

(999, 7)
                      title  ...   country
movie                        ...          
10001                시네마 천국  ...  이탈리아,프랑스
10002              빽 투 더 퓨쳐  ...        미국
10003            빽 투 더 퓨쳐 2  ...        미국
10004            빽 투 더 퓨쳐 3  ...        미국
10005  스타워즈 에피소드 4 - 새로운 희망  ...        미국

[5 rows x 7 columns]

movies_df.columns

Index(['title', 'title_eng', 'year', 'grade', 'genres', 'people', 'country'], dtype='object')

factorization machine에는 feature vector가 있다.
feature vector : user onehot vector + item onehot vector + meta information + other feature engineered vectors

movies_df['genres'].head()

movie
     드라마/멜로/로맨스
         SF/코미디
         SF/코미디
  서부/SF/판타지/코미디
   판타지/모험/SF/액션
Name: genres, dtype: object

# genre onehot vector (genre feature engineering)
dummy_genres_df = movies_df['genres'].str.get_dummies(sep='/')

dummy_genres_df.head()

	SF	가족	공포	느와르	다큐멘터리	드라마	로맨스	멜로	모험	뮤지컬	미스터리	범죄	서부	서사	스릴러	애니메이션	액션	에로	전쟁	코미디	판타지
movie
10001	0	0	0	0	0	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
10002	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
10003	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
10004	1	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	1	1
10005	1	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	1	0	0	0	1

movies_df['grade'].unique()

array(['전체 관람가', '12세 관람가', 'PG', '15세 관람가', 'NR', '청소년 관람불가', 'PG-13',
       'R', 'G', nan], dtype=object)

dummy_grade_df = pd.get_dummies(movies_df['grade'], prefix='grade')
dummy_grade_df.head()

	grade_12세 관람가	grade_15세 관람가	grade_G	grade_NR	grade_PG	grade_PG-13	grade_R	grade_전체 관람가	grade_청소년 관람불가
movie
10001	0	0	0	0	0	0	0	1	0
10002	1	0	0	0	0	0	0	0	0
10003	1	0	0	0	0	0	0	0	0
10004	0	0	0	0	0	0	0	1	0
10005	0	0	0	0	1	0	0	0	0

Convert To Factorization Machine format

user를 one-hot vector로 나타낸다
item을 one-hot vector로 나타낸다
movies_df에서 categorical features를 만든다

train_df.head()
# (user, movie, rate) -> 2D rating matrix -> matrix factorization, user-based, item-based CF

	user	movie	rate	time
137023	48423	10764	10	1212241560
92868	17307	10170	10	1122185220
94390	18180	10048	10	1573403460
22289	1498	10001	9	1432684500
80155	12541	10022	10	1370458140

train_df = train_df[:1000] # 시간관계상 자르기~
print(train_df.shape)

(1000, 4)

train_df['movie'].apply(lambda x: dummy_genres_df.loc[x])

	SF	가족	공포	느와르	다큐멘터리	드라마	로맨스	멜로	모험	뮤지컬	미스터리	범죄	서부	서사	스릴러	애니메이션	액션	에로	전쟁	코미디	판타지
137023	0	0	0	0	0	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
92868	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0
94390	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
22289	0	0	0	0	0	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
80155	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
2870	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0
120892	1	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	1	0	0	0	0
93371	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
58284	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
4595	0	0	0	0	0	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0

1000 rows × 21 columns

# user의 원 핫 벡터 테스트해보자 
test_df = pd.get_dummies(train_df['user'], prefix='user')
test_df.head()

	...	user_48423
137023	...	1
92868	...	0
94390	...	0
22289	...	0
80155	...	0

5 rows × 924 columns

# movie(item)의 원 핫 벡터 테스트해보자
test_df = pd.get_dummies(train_df['movie'], prefix='movie')
print(test_df.head())

        movie_10001  movie_10002  ...  movie_10988  movie_10998
137023            0            0  ...            0            0
92868             0            0  ...            0            0
94390             0            0  ...            0            0
22289             1            0  ...            0            0
80155             0            0  ...            0            0

[5 rows x 263 columns]

# Xtrain 만들기 전에 테스트해보자 
X_train = pd.concat([pd.get_dummies(train_df['user'], prefix='user'), # user의 원핫 벡터 
           pd.get_dummies(train_df['movie'], prefix='movie'), # item의 원핫벡터
           train_df['movie'].apply(lambda x: dummy_genres_df.loc[x]),# metadata 
           train_df['movie'].apply(lambda x: dummy_grade_df.loc[x])], axis=1) # metadata

X_train.shape

(1000, 1217)

X_train.head(2)

	user_0	user_3	user_4	user_8	user_19	user_25	user_28	user_29	user_41	user_42	user_44	user_45	user_58	user_65	user_67	user_70	user_73	user_79	user_82	user_86	user_94	user_95	user_98	user_107	user_110	user_127	user_130	user_136	user_140	user_161	user_176	user_179	user_181	user_182	user_213	user_214	user_217	user_222	user_224	user_261	...	movie_10953	movie_10955	movie_10962	movie_10965	movie_10970	movie_10971	movie_10980	movie_10981	movie_10988	movie_10998	SF	가족	공포	느와르	다큐멘터리	드라마	로맨스	멜로	모험	뮤지컬	미스터리	범죄	서부	서사	스릴러	애니메이션	액션	에로	전쟁	코미디	판타지	grade_12세 관람가	grade_15세 관람가	grade_G	grade_NR	grade_PG	grade_PG-13	grade_R	grade_전체 관람가	grade_청소년 관람불가
137023	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0
92868	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	1	0	0	0	0

2 rows × 1217 columns

X_train = pd.concat([pd.get_dummies(train_df['user'], prefix='user'),
           pd.get_dummies(train_df['movie'], prefix='movie'),
           train_df['movie'].apply(lambda x: dummy_genres_df.loc[x]),
           train_df['movie'].apply(lambda x: dummy_grade_df.loc[x])], axis=1)
# 평균 평점이 높기 때문에 10만 1로 주고, 나머지는 -1로 binary 지정한다(binary classification)
# loss 계산을 0이 아닌 -1로  지정한다
y_train = train_df['rate'].apply(lambda x: 1 if x > 9 else -1)
print(X_train.shape)
print(y_train.shape)

# 0이 아닌 데이터 위치 확인하기위해 csr matrix 활용한다
# csr_matrix 설명 참고: https://rfriend.tistory.com/551
X_train_sparse = scipy.sparse.csr_matrix(X_train.values)

(1000, 1217)
(1000,)

X_train_sparse # 1000*1217 행렬에서 5614개의 값밖에 없다 (sparse)

<1000x1217 sparse matrix of type '<class 'numpy.longlong'>'
	with 5614 stored elements in Compressed Sparse Row format>

Train Factorization Machine

Model Equation

Pairwise Interaction to be computed
- Linear Complexity

# Compute negative log likelihood between prediction and label
def log_loss(pred, y):
    return np.log(np.exp(-pred * y) + 1.0)

# Update gradients
def sgd(X, y, n_samples, n_features,
                w0, w, v, n_factors, learning_rate, reg_w, reg_v):
    data = X.data
    indptr = X.indptr # indptr[i]는 i번째 행의 원소가 data의 어느 인덱스에서 시작되는지 나타낸다
    indices = X.indices # indices[i]는 data[i]의 열 번호를 저장한다. 
    loss = 0.0

    for i in range(n_samples):
        pred, summed = predict(X, w0, w, v, n_factors, i)
        
        # calculate loss and its gradient
        loss += log_loss(pred, y[i])
        loss_gradient = -y[i] / (np.exp(y[i] * pred) + 1.0)
    
        # update bias/intercept term
        w0 -= learning_rate * loss_gradient

        # update weight
        for index in range(indptr[i], indptr[i + 1]):
            feature = indices[index]
            w[feature] -= learning_rate * (loss_gradient * data[index] + 2 * reg_w * w[feature])

        # update factor
        for factor in range(n_factors):
            for index in range(indptr[i], indptr[i + 1]):
                feature = indices[index]
                term = summed[factor] - v[factor, feature] * data[index]
                v_gradient = loss_gradient * data[index] * term
                v[factor, feature] -= learning_rate * (v_gradient + 2 * reg_v * v[factor, feature])
    
    loss /= n_samples
    return loss

def predict(X, w0, w, v, n_factors, i):
    data = X.data
    indptr = X.indptr
    indices = X.indices
    """predicting a single instance"""
    summed = np.zeros(n_factors)
    summed_squared = np.zeros(n_factors)

    # linear output w * x
    pred = w0
    for index in range(indptr[i], indptr[i + 1]):
        feature = indices[index]
        pred += w[feature] * data[index]

    # factor output
    for factor in range(n_factors):
        for index in range(indptr[i], indptr[i + 1]):
            feature = indices[index]
            term = v[factor, feature] * data[index]
            summed[factor] += term
            summed_squared[factor] += term * term

        pred += 0.5 * (summed[factor] * summed[factor] - summed_squared[factor])

    # gradient update할 때, summed는 독립이므로 re-use 가능
    return pred, summed

# Train Factorization Machine
# X -> sparse csr_matrix, y -> label
def fit(X, y, config):
    epochs = config['num_epochs']
    num_factors = config['num_factors']
    learning_rate = config['learning_rate']
    reg_weights = config['reg_weights']
    reg_features = config['reg_features']

    num_samples, num_features = X.shape
    weights = np.zeros(num_features) # -> w
    global_bias = 0.0 # -> w0
    
    # latent factors for all features -> v
    feature_factors = np.random.normal(size = (num_factors, num_features))

    epoch_loss = []
    for epoch in range(epochs):
        loss = sgd(X, y, num_samples, num_features,
                            global_bias, weights,
                            feature_factors, num_factors,
                            learning_rate, reg_weights, reg_features)
        print(f'[epoch: {epoch+1}], loss: {loss}')

        epoch_loss.append(loss)
      
    return epoch_loss

config = {
    "num_epochs": 10,
    "num_factors": 10,
    "learning_rate": 0.1,
    "reg_weights": 0.01,
    "reg_features": 0.01
}

epoch_loss = fit(X_train_sparse, y_train.values, config)

[epoch: 1], loss: 2.5530346045055703
[epoch: 2], loss: 1.0115441271017367
[epoch: 3], loss: 0.4945105895776768
[epoch: 4], loss: 0.26980520496116406
[epoch: 5], loss: 0.176137422897979
[epoch: 6], loss: 0.1321992253904276
[epoch: 7], loss: 0.10100276394910408
[epoch: 8], loss: 0.08020458255299875
[epoch: 9], loss: 0.06571114718797659
[epoch: 10], loss: 0.05607737598425833

import matplotlib.pyplot as plt
plt.plot(epoch_loss)
plt.title('Loss per epoch')
plt.show()

png

Twitter Facebook LinkedIn

서다솔

[논문구현]Factorization Machines(2010)

Factorization Machine

Configuration

Data Loader

rates.csv

movie dataframe

Convert To Factorization Machine format

Train Factorization Machine

공유하기

참고

[보안] 타사 인증서 AWS ALB SSL 적용

속도 개선을 위한 DB 인덱싱 및 비동기 적용

[AWS] API Gateway 호출 기록 로깅

[CICD] git-jenkins-ECR로 자동배포 구축하기