13. Cross validation & Pipeline

0. 개요

교차 검증과 파이프라인에 대해서 학습

 

---

 

교차검증

### Building a Pipeline ###

# load iris and create X and y

import pandas as pd

import seaborn as sns
sns.set_context("paper", font_scale = 1.5)
sns.set_style("white")


from sklearn.datasets import load_iris

dataset = load_iris()
X, y = dataset.data, dataset.target


from sklearn.decomposition import PCA

pca = PCA(n_components = 2)

# fit and transform using 2 input dimensions
out_pca = pca.fit_transform(X)

# create pca output dataframe and add label column "species" 
df_pca = pd.DataFrame(data = out_pca, columns = ['pca1', 'pca2'])

df_pca = pd.concat([df_pca, pd.DataFrame(y, columns = ['species'])], axis = 1)

# plot scatter of pca data
sns.lmplot(x = 'pca1', y = 'pca2', hue = 'species', data = df_pca, fit_reg = False)


sns.violinplot(x = 'species', y = 'pca1', data = df_pca)

sns.violinplot(x = 'species', y = 'pca2', data = df_pca)



# 로지스틱 회귀 & f1 스코어 
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import f1_score

# create train and test sets
X_train, X_test, y_train, y_test = \
        train_test_split(X, y, test_size = 0.30)

# instantiate classifier object and fit to training data
clf = LogisticRegression()
clf.fit(X_train, y_train)

# predict on test set and score the predictions against y_test
y_pred = clf.predict(X_test)
f1 = f1_score(y_test, y_pred, average = 'weighted') 
print('f1 score is = ' + str(f1))



# PCA 거친 데이터로 분석하기
X_train, X_test, y_train, y_test = \
        train_test_split(df_pca[['pca1', 'pca2']], df_pca['species'], test_size = 0.30)


clf = LogisticRegression()
clf.fit(X_train, y_train)

# predict on test set and score the predictions against y_test
y_pred = clf.predict(X_test)
f1 = f1_score(y_test, y_pred, average = 'weighted') 
print('f1 score is = ' + str(f1))




pca = PCA(n_components = 3)

# fit and transform using 2 input dimensions
out_pca = pca.fit_transform(X)

# create pca output dataframe and add label column "species" 
df_pca = pd.DataFrame(data = out_pca, columns = ['pca1', 'pca2', 'pca3'])

df_pca = pd.concat([df_pca, pd.DataFrame(y, columns = ['species'])], axis = 1)

sns.violinplot(x = 'species', y = 'pca1', data = df_pca)

sns.violinplot(x = 'species', y = 'pca2', data = df_pca)

sns.violinplot(x = 'species', y = 'pca3', data = df_pca)


X_train, X_test, y_train, y_test = \
        train_test_split(df_pca[['pca1', 'pca2', 'pca3']], df_pca['species'], test_size = 0.30)


clf = LogisticRegression()
clf.fit(X_train, y_train)

# predict on test set and score the predictions against y_test
y_pred = clf.predict(X_test)
f1 = f1_score(y_test, y_pred, average = 'weighted') 
print('f1 score is = ' + str(f1))

 

 

 

파이프라인

## Pipeline ##
# import modules 

from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV

# create train and test sets
X_train, X_test, y_train, y_test = \
        train_test_split(X, y, test_size = 0.30)
        
pca = PCA()

logistic = LogisticRegression()

# instantiate a pipeline and add steps to the pipeline
pipe = Pipeline(steps = [('pca', pca), ('logistic', logistic)])

# print list of steps with names
print(pipe.steps[0])


# set the parameter grid to be passed to the grid search
param_grid = {
    'pca__n_components': [2, 3, 4]
}


# instantiate the grid search object and pass the pipe and param_grid
model = GridSearchCV(pipe, param_grid, iid = False, cv = 5,
                      return_train_score = False)


# fit entire pipeline using grid serach and 5-fold cross validation
model.fit(X_train, y_train)

print("Best parameter (CV score = %0.3f):" % model.best_score_)
print(model.best_params_)

y_pred = model.predict(X_test)

 

 

 

파이프라인 응용 예제

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeClassifier
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV

# 데이터 스플릿
X_train, X_test, y_train, y_test = train_test_split(cancer.data, cancer.target, 
                                                    test_size = 0.1,
                                                    stratify = cancer.target, 
                                                    random_state = 42)

# 모델 객체 생성
tree = DecisionTreeClassifier(criterion = 'gini', max_depth = 1, random_state = 0)

# 파이프라인 설정
pipe = Pipeline(steps = [('tree', tree)])

###############################################################
# 반복할 변수 설정
param_grid = {
    'tree__max_depth': [2, 4, 5,6,7]
}

# * 트리는 max_depth로 깊이를 설정하므로 해당 파라미터의 값에 들어갈 반복에 대한 설정이다. 
# 매우 중요한 부분
###############################################################


# 파이프라인 모델 적합, 옵션은 5겹
model = GridSearchCV(pipe, param_grid, iid = False, cv = 5,
                      return_train_score = False)


# 파이프라인 모델 적합
model.fit(X_train, y_train)

# 결과 출력
print("Best parameter (CV score = %0.3f):" % model.best_score_)
print(model.best_params_)

y_pred = model.predict(X_test)