機(jī)器學(xué)習(xí)基礎(chǔ)：缺失值的處理技巧（附Python代碼）

import pandas as pd
# 統(tǒng)計(jì)缺失值數(shù)量missing=data.isnull().sum().reset_index().rename(columns={0:'missNum'})# 計(jì)算缺失比例missing['missRate']=missing['missNum']/data.shape[0]# 按照缺失率排序顯示miss_analy=missing[missing.missRate>0].sort_values(by='missRate',ascending=False)# miss_analy 存儲(chǔ)的是每個(gè)變量缺失情況的數(shù)據(jù)框

import matplotlib.pyplot as pltimport pylab as pl
fig = plt.figure(figsize=(18,6))plt.bar(np.arange(miss_analy.shape[0]), list(miss_analy.missRate.values), align = 'center',color=['red','green','yellow','steelblue'])
plt.title('Histogram of missing value of variables')plt.xlabel('variables names')plt.ylabel('missing rate')# 添加x軸標(biāo)簽，并旋轉(zhuǎn)90度plt.xticks(np.arange(miss_analy.shape[0]),list(miss_analy['index']))pl.xticks(rotation=90)# 添加數(shù)值顯示for x,y in enumerate(list(miss_analy.missRate.values)):    plt.text(x,y+0.12,'{:.2%}'.format(y),ha='center',rotation=90)    plt.ylim([0,1.2])
plt.show()

func: df.dropna(axis=0, how='any', thresh=None, subset=None, inplace=False)
# 1、刪除‘a(chǎn)ge’列df.drop('age', axis=1, inplace=True)
# 2、刪除數(shù)據(jù)表中含有空值的行df.dropna()
# 3、丟棄某幾列有缺失值的行df.dropna(axis=0, subset=['a','b'], inplace=True)

# 去掉缺失比例大于80%以上的變量data=data.dropna(thresh=len(data)*0.2, axis=1)

# 均值填充data['col'] = data['col'].fillna(data['col'].means())# 中位數(shù)填充data['col'] = data['col'].fillna(data['col'].median())# 眾數(shù)填充data['col'] = data['col'].fillna(stats.mode(data['col'])[0][0])

from sklearn.preprocessing import Imputerimr = Imputer(missing_values='NaN', strategy='mean', axis=0)imputed_data =pd.DataFrame(imr.fit_transform(df.values),columns=df.columns)imputed_data

#  interpolate()插值法，缺失值前后數(shù)值的均值，但是若缺失值前后也存在缺失，則不進(jìn)行計(jì)算插補(bǔ)。df['a'] = df['a'].interpolate()
# 用前面的值替換, 當(dāng)?shù)谝恍杏腥笔е禃r(shí)，該行利用向前替換無值可取，仍缺失df.fillna(method='pad')
# 用后面的值替換，當(dāng)最后一行有缺失值時(shí)，該行利用向后替換無值可取，仍缺失df.fillna(method='backfill')#用后面的值替換

from fancyimpute import KNNfill_knn = KNN(k=3).fit_transform(data)data = pd.DataFrame(fill_knn)

from sklearn.neighbors import KNeighborsClassifier, KNeighborsRegressor
def knn_filled_func(x_train, y_train, test, k = 3, dispersed = True):    # params: x_train 為目標(biāo)列不含缺失值的數(shù)據(jù)（不包括目標(biāo)列）    # params: y_train 為不含缺失值的目標(biāo)列    # params: test 為目標(biāo)列為缺失值的數(shù)據(jù)（不包括目標(biāo)列）    if dispersed:        knn= KNeighborsClassifier(n_neighbors = k, weights = "distance")    else:        knn= KNeighborsRegressor(n_neighbors = k, weights = "distance")
    knn.fit(x_train, y_train)    return test.index, knn.predict(test)

from sklearn.ensemble import RandomForestRegressor, RandomForestClassifier
def knn_filled_func(x_train, y_train, test, k = 3, dispersed = True):    # params: x_train 為目標(biāo)列不含缺失值的數(shù)據(jù)（不包括目標(biāo)列）    # params: y_train 為不含缺失值的目標(biāo)列    # params: test 為目標(biāo)列為缺失值的數(shù)據(jù)（不包括目標(biāo)列）    if dispersed:        rf= RandomForestRegressor()    else:        rf= RandomForestClassifier()
    rf.fit(x_train, y_train)    return test.index, rf.predict(test)