# 生成不平衡分類數(shù)據(jù)集
from collections import Counter
from sklearn.datasets import make_classification
X, y = make_classification(n_samples=3000, n_features=2, n_informative=2,
              n_redundant=0, n_repeated=0, n_classes=3,
              n_clusters_per_class=1,
              weights=[0.1, 0.05, 0.85],
              class_sep=0.8, random_state=2018)
Counter(y)
# Counter({2: 2532, 1: 163, 0: 305})

# 使用RandomOverSampler從少數(shù)類的樣本中進(jìn)行隨機(jī)采樣來(lái)增加新的樣本使各個(gè)分類均衡
from imblearn.over_sampling import RandomOverSampler
 
ros = RandomOverSampler(random_state=0)
X_resampled, y_resampled = ros.fit_sample(X, y)
sorted(Counter(y_resampled).items())
# [(0, 2532), (1, 2532), (2, 2532)]

# SMOTE: 對(duì)于少數(shù)類樣本a, 隨機(jī)選擇一個(gè)最近鄰的樣本b, 然后從a與b的連線上隨機(jī)選取一個(gè)點(diǎn)c作為新的少數(shù)類樣本
from imblearn.over_sampling import SMOTE
 
X_resampled_smote, y_resampled_smote = SMOTE().fit_sample(X, y)
 
sorted(Counter(y_resampled_smote).items())
# [(0, 2532), (1, 2532), (2, 2532)]

# ADASYN: 關(guān)注的是在那些基于K最近鄰分類器被錯(cuò)誤分類的原始樣本附近生成新的少數(shù)類樣本
from imblearn.over_sampling import ADASYN

X_resampled_adasyn, y_resampled_adasyn = ADASYN().fit_sample(X, y)
 
sorted(Counter(y_resampled_adasyn).items())
# [(0, 2522), (1, 2520), (2, 2532)]

# RandomUnderSampler函數(shù)是一種快速并十分簡(jiǎn)單的方式來(lái)平衡各個(gè)類別的數(shù)據(jù): 隨機(jī)選取數(shù)據(jù)的子集.
from imblearn.under_sampling import RandomUnderSampler
rus = RandomUnderSampler(random_state=0)
X_resampled, y_resampled = rus.fit_sample(X, y)
 
sorted(Counter(y_resampled).items())
# [(0, 163), (1, 163), (2, 163)]

# 在之前的SMOTE方法中, 當(dāng)由邊界的樣本與其他樣本進(jìn)行過(guò)采樣差值時(shí), 很容易生成一些噪音數(shù)據(jù). 因此, 在過(guò)采樣之后需要對(duì)樣本進(jìn)行清洗. 
# 這樣TomekLink 與 EditedNearestNeighbours方法就能實(shí)現(xiàn)上述的要求.
from imblearn.combine import SMOTEENN
smote_enn = SMOTEENN(random_state=0)
X_resampled, y_resampled = smote_enn.fit_sample(X, y)
 
sorted(Counter(y_resampled).items())
# [(0, 2111), (1, 2099), (2, 1893)]

from imblearn.combine import SMOTETomek
smote_tomek = SMOTETomek(random_state=0)
X_resampled, y_resampled = smote_tomek.fit_sample(X, y)
 
sorted(Counter(y_resampled).items())
# [(0, 2412), (1, 2414), (2, 2396)]

# 使用SVM的權(quán)重調(diào)節(jié)處理不均衡樣本 權(quán)重為balanced 意味著權(quán)重為各分類數(shù)據(jù)量的反比
from sklearn.svm import SVC 
svm_model = SVC(class_weight='balanced')
svm_model.fit(X, y)

# # EasyEnsemble 通過(guò)對(duì)原始的數(shù)據(jù)集進(jìn)行隨機(jī)下采樣實(shí)現(xiàn)對(duì)數(shù)據(jù)集進(jìn)行集成.
# EasyEnsemble 有兩個(gè)很重要的參數(shù): (i) n_subsets 控制的是子集的個(gè)數(shù) and (ii) replacement 決定是有放回還是無(wú)放回的隨機(jī)采樣.
from imblearn.ensemble import EasyEnsemble
ee = EasyEnsemble(random_state=0, n_subsets=10)
X_resampled, y_resampled = ee.fit_sample(X, y)
sorted(Counter(y_resampled[0]).items())
# [(0, 163), (1, 163), (2, 163)]

# BalanceCascade(級(jí)聯(lián)平衡)的方法通過(guò)使用分類器(estimator參數(shù))來(lái)確保那些被錯(cuò)分類的樣本在下一次進(jìn)行子集選取的時(shí)候也能被采樣到. 同樣, n_max_subset 參數(shù)控制子集的個(gè)數(shù), 以及可以通過(guò)設(shè)置bootstrap=True來(lái)使用bootstraping(自助法).
from imblearn.ensemble import BalanceCascade
from sklearn.linear_model import LogisticRegression
bc = BalanceCascade(random_state=0,
          estimator=LogisticRegression(random_state=0),
          n_max_subset=4)
X_resampled, y_resampled = bc.fit_sample(X, y)
 
sorted(Counter(y_resampled[0]).items())
# [(0, 163), (1, 163), (2, 163)]

# BalancedBaggingClassifier 允許在訓(xùn)練每個(gè)基學(xué)習(xí)器之前對(duì)每個(gè)子集進(jìn)行重抽樣. 簡(jiǎn)而言之, 該方法結(jié)合了EasyEnsemble采樣器與分類器(如BaggingClassifier)的結(jié)果.
from sklearn.tree import DecisionTreeClassifier
from imblearn.ensemble import BalancedBaggingClassifier
bbc = BalancedBaggingClassifier(base_estimator=DecisionTreeClassifier(),
                ratio='auto',
                replacement=False,
                random_state=0)
bbc.fit(X, y)