Sklearn Classifier

 

 

 

Sklearn에서 제공하는 Classifier 모듈을 활용하여 분류 모델을 만듭니다.

Task는 긍정, 중립, 부정 3개로 분류하는 감성 분류 모델입니다.

 

 

 

sklearn 주요 모듈

 

 

 

 

 

👉🏻 Data Load & Check

vectorize는 content 데이터로 TF-IDF vectorizing 한 'vec'을 활용하였습니다.

 

df.head()

 

 

 

👉🏻 Train Test Split

 

from sklearn.model_selection import train_test_split

train_data, test_data = train_test_split(df, test_size=0.2, random_state=42)
train_data = train_data.reset_index()
test_data = test_data.reset_index()

print(len(df))
print(len(train_data))
print(len(test_data))

 

 

👉🏻 Classfier

Sklearn 제공 분류 모델을 모두 활용하여 비교합니다.

 

# kNN
from sklearn.neighbors import KNeighborsClassifier
# Decision Tree
from sklearn.tree import DecisionTreeClassifier
# Random Forest
from sklearn.ensemble import RandomForestClassifier
# Naive Bayes
from sklearn.naive_bayes import GaussianNB
# SVM, Support Vector Machines
from sklearn.svm import SVC

from sklearn.model_selection import KFold, cross_val_score

 

Score

kfold, cross_val_score로 훈련 테스트를 바꿔가며 train하는 교차검증으로 정확도를 비교합니다.

 

scoring = 'accuracy'
k_fold = KFold(shuffle = True, random_state=0)

# knn
knn_clf = KNeighborsClassifier()
knn_clf = knn_clf.fit(train_vec, train_label)
knn_score = cross_val_score(knn_clf, train_vec, train_label, cv=k_fold, n_jobs=1, scoring=scoring)

# decision tree
dt_clf = DecisionTreeClassifier()
dt_clf = dt_clf.fit(train_vec, train_label)
dt_score = cross_val_score(dt_clf, train_vec, train_label, cv=k_fold, n_jobs=1, scoring=scoring)

# random forest
rf_clf = RandomForestClassifier()
rf_clf = rf_clf.fit(train_vec, train_label)
rf_score = cross_val_score(rf_clf, train_vec, train_label, cv=k_fold, n_jobs=1, scoring=scoring)

# naive bayes
nb_clf = GaussianNB()
nb_clf = nb_clf.fit(train_vec, train_label)
nb_score = cross_val_score(nb_clf, train_vec, train_label, cv=k_fold, n_jobs=1, scoring=scoring)

# svm
svm_clf = SVC()
svm_clf = svm_clf.fit(train_vec, train_label)
svm_score = cross_val_score(svm_clf, train_vec, train_label, cv=k_fold, n_jobs=1, scoring=scoring)

print(f"kNN_score : {round(np.mean(knn_score) * 100, 2)}")
print(f"DecisionTree_score : {round(np.mean(dt_score) * 100, 2)}")
print(f"RandomForest_score : {round(np.mean(rf_score) * 100, 2)}")
print(f"naivebayes_score : {round(np.mean(nb_score) * 100, 2)}")
print(f"svm_score : {round(np.mean(svm_score) * 100, 2)}")

RandomForest model이 score가 가장 높으므로 채택.

 

 

 

👉🏻 RandomForest

 

from sklearn.ensemble import RandomForestClassifier

rf_clf = RandomForestClassifier()
rf_clf.fit(train_vec, train_label)

# labeling
test_pred = rf_clf.predict(test_vec)
test_data['label'] = test_pred

test_data.head()

 

 

metrics

평가지표로 f1 score 채택.

 

from sklearn.metrics import f1_score
f1_score(test_data.sentiment, test_pred, average='micro')

 

 

 

👉🏻 GridSearchCV

GridSearch를 통해 좀 더 최적의 하이퍼파라미터를 튜닝해보겠습니다.

 

from sklearn.metrics import classification_report, f1_score, accuracy_score
from sklearn.model_selection import GridSearchCV

params = { 'n_estimators' : [100, 200],
           'max_depth' : [6, 8, 10, 12],
           'min_samples_leaf' : [8, 12, 18],
           'min_samples_split' : [8, 16, 20]
            }

# RandomForestClassifier 객체 생성 후 GridSearchCV 수행
rf_clf = RandomForestClassifier(random_state = 0, n_jobs = -1)
grid_cv = GridSearchCV(rf_clf,
                       param_grid = params,
                       cv = 3,
                       n_jobs = -1)
grid_cv.fit(train_vec, train_label)

print('최적 하이퍼 파라미터: ', grid_cv.best_params_)
print('최고 예측 정확도: {:.4f}'.format(grid_cv.best_score_))

 

 

최적의 하이퍼 파라미터로 재학습

 

rf_clf = RandomForestClassifier(n_estimators = 200,          # 결정 트리 개수
                                 max_depth = 12,            # 트리의 최대 깊이
                                 min_samples_leaf = 8,      # 리프노드가 되기 위해 필요한 최소한의 샘플 데이터수 
                                 min_samples_split = 8,    # 노드를 분할하기 위한 최소한의 샘플 데이터수(과적합을 제어하는데 사용)
                                 random_state = 0,
                                 n_jobs = -1)

rf_clf.fit(train_vec, train_label)

# labeling
test_pred = rf_clf.predict(test_vec)
test_data['label'] = test_pred

test_data.head()

 

f1_score(test_data.sentiment, test_pred, average='micro')