0. 핵심개념 및 사이킷런 알고리즘 API 링크
sklearn.neural_network.MLPClassifier
sklearn.neural_network.MLPRegressor
Part1. 분류(Classification)
1. 분석 데이터 준비
import warnings
warnings.filterwarnings("ignore")
import pandas as pd
data1=pd.read_csv('breast-cancer-wisconsin.csv', encoding='utf-8')
X=data1[data1.columns[1:10]]
y=data1[["Class"]]
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test=train_test_split(X, y, stratify=y, random_state=42)
from sklearn.preprocessing import MinMaxScaler
scaler=MinMaxScaler()
scaler.fit(X_train)
X_scaled_train=scaler.transform(X_train)
X_scaled_test=scaler.transform(X_test)2. 기본모델 적용
from sklearn.neural_network import MLPClassifier
model=MLPClassifier()
model.fit(X_scaled_train, y_train)
pred_train=model.predict(X_scaled_train)
model.score(X_scaled_train, y_train)
0.97265625
from sklearn.metrics import confusion_matrix
confusion_train=confusion_matrix(y_train, pred_train)
print("훈련데이터 오차행렬:\n", confusion_train)
훈련데이터 오차행렬:
[[327 6]
[ 8 171]]
from sklearn.metrics import classification_report
cfreport_train=classification_report(y_train, pred_train)
print("분류예측 레포트:\n", cfreport_train)
분류예측 레포트:
precision recall f1-score support
0 0.98 0.98 0.98 333
1 0.97 0.96 0.96 179
accuracy 0.97 512
macro avg 0.97 0.97 0.97 512
weighted avg 0.97 0.97 0.97 512
pred_test=model.predict(X_scaled_test)
model.score(X_scaled_test, y_test)
0.9590643274853801
confusion_test=confusion_matrix(y_test, pred_test)
print("테스트데이터 오차행렬:\n", confusion_test)
테스트데이터 오차행렬:
[[106 5]
[ 2 58]]
from sklearn.metrics import classification_report
cfreport_test=classification_report(y_test, pred_test)
print("분류예측 레포트:\n", cfreport_test)
분류예측 레포트:
precision recall f1-score support
0 0.98 0.95 0.97 111
1 0.92 0.97 0.94 60
accuracy 0.96 171
macro avg 0.95 0.96 0.96 171
weighted avg 0.96 0.96 0.96 171
3. Grid Search
param_grid={'hidden_layer_sizes': [10, 30, 50, 100], 'solver': ['sgd', 'adam'],
'activation': ['tanh', 'relu']}
from sklearn.model_selection import GridSearchCV
grid_search=GridSearchCV(MLPClassifier(), param_grid, cv=5)
grid_search.fit(X_scaled_train, y_train)
GridSearchCV(cv=5, estimator=MLPClassifier(),
param_grid={'activation': ['tanh', 'relu'],
'hidden_layer_sizes': [10, 30, 50, 100],
'solver': ['sgd', 'adam']})
print("Best Parameter: {}".format(grid_search.best_params_))
print("Best Score: {:.4f}".format(grid_search.best_score_))
print("TestSet Score: {:.4f}".format(grid_search.score(X_scaled_test, y_test)))
Best Parameter: {'activation': 'tanh', 'hidden_layer_sizes': 30, 'solver': 'adam'}
Best Score: 0.9785
TestSet Score: 0.95914. Random Search
from scipy.stats import randint
param_distribs={'hidden_layer_sizes': randint(low=10, high=100), 'solver': ['sgd', 'adam'],
'activation': ['tanh', 'relu']}
from sklearn.model_selection import RandomizedSearchCV
random_search=RandomizedSearchCV(MLPClassifier(),
param_distributions=param_distribs, n_iter=10, cv=5)
random_search.fit(X_scaled_train, y_train)
RandomizedSearchCV(cv=5, estimator=MLPClassifier(),
param_distributions={'activation': ['tanh', 'relu'],
'hidden_layer_sizes': <scipy.stats._distn_infrastructure.rv_frozen object at 0x00000187DF946FD0>,
'solver': ['sgd', 'adam']})
print("Best Parameter: {}".format(random_search.best_params_))
print("Best Score: {:.4f}".format(random_search.best_score_))
print("TestSet Score: {:.4f}".format(random_search.score(X_scaled_test, y_test)))
Best Parameter: {'activation': 'relu', 'hidden_layer_sizes': 68, 'solver': 'adam'}
Best Score: 0.9726
TestSet Score: 0.9591Part2. 회귀(Regression)
1. 분석 데이터 준비
data2=pd.read_csv('house_price.csv', encoding='utf-8')
X=data2[data2.columns[1:5]]
y=data2[["house_value"]]
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test=train_test_split(X, y, random_state=42)
from sklearn.preprocessing import MinMaxScaler
scaler=MinMaxScaler()
scaler.fit(X_train)
X_scaled_train=scaler.transform(X_train)
X_scaled_test=scaler.transform(X_test)2. 기본모델 적용
from sklearn.neural_network import MLPRegressor
model=MLPRegressor()
model.fit(X_scaled_train, y_train)
pred_train=model.predict(X_scaled_train)
model.score(X_scaled_train, y_train)
-2.8006239432716575
pred_test=model.predict(X_scaled_test)
model.score(X_scaled_test, y_test)
-2.758166270245026
# RMSE (Root Mean Squared Error)
import numpy as np
from sklearn.metrics import mean_squared_error
MSE_train = mean_squared_error(y_train, pred_train)
MSE_test = mean_squared_error(y_test, pred_test)
print("훈련 데이터 RMSE:", np.sqrt(MSE_train))
print("테스트 데이터 RMSE:", np.sqrt(MSE_test))
훈련 데이터 RMSE: 186070.9152873358
테스트 데이터 RMSE: 185332.024186712863. 튜닝 모델
from sklearn.neural_network import MLPRegressor
model=MLPRegressor(hidden_layer_sizes=(64,64,64),activation="relu" ,random_state=1, max_iter=2000)
model.fit(X_scaled_train, y_train)
pred_train=model.predict(X_scaled_train)
model.score(X_scaled_train, y_train)
0.5661979037463138
pred_test=model.predict(X_scaled_test)
model.score(X_scaled_test, y_test)
0.5840866843135079
# RMSE (Root Mean Squared Error)
import numpy as np
from sklearn.metrics import mean_squared_error
MSE_train = mean_squared_error(y_train, pred_train)
MSE_test = mean_squared_error(y_test, pred_test)
print("훈련 데이터 RMSE:", np.sqrt(MSE_train))
print("테스트 데이터 RMSE:", np.sqrt(MSE_test))
훈련 데이터 RMSE: 62863.2553580582
테스트 데이터 RMSE: 61654.37310884089