카테고리형(명목형) 변수를 가진 데이터를 딥러닝으로 풀어보자 (with Bank Marketing Data Set)

import numpy as np
import pandas as pd
import tensorflow as tf

np.set_printoptions(precision=3)

from tensorflow.keras.utils import get_file

fpath = get_file("bank-full.csv",
                 "https://docs.google.com/uc?id=16Z2Jyg9BPB8kLeuGDRNLpXZdF77W32_p")
df = pd.read_csv(fpath, sep=';')

print("데이터의 크기 : {df.shape}")
df.head()

cat_col_names = ["marital",'job','contact','education','month',"poutcome","housing","loan",'default']

for col_name in cat_col_names:
    cat_values = np.unique(df[col_name])
    print(f"{col_name}: {cat_values}\n")

num_col_names = ['age', 'balance', 'day', 'duration',
                 'campaign','pdays', 'previous']

for col_name in num_col_names:
    print(f"{col_name}: ({df[col_name].min()},{df[col_name].max()})")

from sklearn.preprocessing import LabelEncoder

category_xs = []
category_encoders = []
for col_name in cat_col_names:
    encoder = LabelEncoder()
    encoded_xs = encoder.fit_transform(df[col_name])

    category_xs.append(encoded_xs)
    category_encoders.append(encoder)

category_xs = np.stack(category_xs, axis=1) 
category_xs

from sklearn.preprocessing import StandardScaler

numeric_encoder = StandardScaler()
numeric_xs = numeric_encoder.fit_transform(df[num_col_names])
numeric_xs

ys = df.y.map({'yes':True,'no':False}).values
ys

from sklearn.model_selection import train_test_split

splitted = train_test_split(category_xs, numeric_xs, ys, 
                            test_size=0.1,stratify=ys)

train_category_xs, train_numeric_xs, train_ys = splitted[::2]
test_category_xs, test_numeric_xs, test_ys = splitted[1::2]

from tensorflow.keras.layers import Input

category_inputs = Input((9,), dtype=tf.int32)
numeric_inputs = Input((7,), dtype=tf.float32)

from tensorflow.keras.layers import Embedding

# 변수 별 임베딩 크기
embed_size = 4

embeds = []
for idx, col_name in enumerate(cat_col_names):
    # 각 변수 별 카테고리의 갯수
    category_size = len(category_encoders[idx].classes_)
    
    # 각 변수 별로 임베딩 레이어 적용 (-> 각 변수 내 카테고리별로 임베딩 값들이 모델에서 학습)
    category_embeded = Embedding(
        category_size, embed_size, name=col_name+'_embed')(category_inputs[:,idx])
    
    embeds.append(category_embeded)
embeds    

from tensorflow.keras.layers import Concatenate

inputs_list = embeds + [numeric_inputs]
concats = Concatenate(name='embed_concat')(inputs_list)
concats

from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Dropout
from tensorflow.keras.regularizers import l2
from tensorflow.keras.models import Model

hidden = Dense(50,activation='relu', name='hidden1',
               kernel_regularizer=l2(1e-5))(concats)
hidden = Dropout(0.3, name='dropout1')(hidden)
hidden = Dense(50,activation='relu', name='hidden2',
               kernel_regularizer=l2(1e-5))(hidden)
hidden = Dropout(0.3, name='dropout2')(hidden)
output = Dense(1, activation='sigmoid', name='output',
               kernel_regularizer=l2(1e-5))(hidden)

model = Model([category_inputs, numeric_inputs], output)

from tensorflow.keras.optimizers import Adagrad
from tensorflow.keras.losses import BinaryCrossentropy
from tensorflow.keras.metrics import BinaryAccuracy

model.compile(optimizer=Adagrad(1e-2),
              loss=BinaryCrossentropy(),
              metrics=[BinaryAccuracy()])

train_xs = [train_category_xs,train_numeric_xs]
model.fit(x=train_xs, y=train_ys,
          batch_size=64, epochs=100, 
          validation_split=0.1, verbose=0);

loss, acc = model.evaluate(x=[test_category_xs, test_numeric_xs], y=test_ys, verbose=0)
print(f"딥러닝 모형의 정확도 : {acc:.3%}")

from sklearn.ensemble import RandomForestClassifier

rf_clf = RandomForestClassifier()

# 모형 학습시키기
train_xs = np.concatenate([train_category_xs,train_numeric_xs],axis=1)
rf_clf.fit(train_xs, train_ys)

# 모형 평가하기 
test_xs = np.concatenate([test_category_xs,test_numeric_xs],axis=1)
rf_clf.score(test_xs, test_ys)