1）讀取csv文件

data =pandas.read_csv(‘test.csv’) //返回的是DataFrame變量

first_rows = data.head(n) //返回前n條數(shù)據(jù),默認(rèn)返回5條

cols = data.columns //返回全部列名

dimensison = data.shape //返回?cái)?shù)據(jù)的格式，數(shù)組，（行數(shù)，列數(shù)）

data.values //返回底層的numpy數(shù)據(jù)

如下去所示的csv數(shù)據(jù)：

解析1:

import pandas as pdtrain_data = pd.read_csv("train.csv")# 將標(biāo)簽轉(zhuǎn)為0，1，2，3，4，...# 去掉重復(fù)的species = train_data['species'].unique()PRint species# 轉(zhuǎn)為0，1，2，3，4，...species = pd.Series(range(0, len(species)), index=list(species))print species# 修改標(biāo)簽數(shù)據(jù)為：0，1，2，3，4，...def fix_spieces(line):    line['species'] = species.loc[line['species']]    return linetrain_data = train_data.apply(fix_spieces, axis=1)print train_data[0:4]# 返回numpy數(shù)據(jù)。# 下標(biāo)從0開始。# 取樣本數(shù)據(jù)，所有行：第二列到最后一列。train_data.values[0::, 2::]# 取標(biāo)簽數(shù)據(jù)，所有行：第一列。train_data.values[0::, 1]
解析2:
import pandas as pdimport numpy as npfrom sklearn.preprocessing import LabelEncoderfrom sklearn.preprocessing import StandardScalertrain_data = pd.read_csv("train.csv")# 將train_data中的‘id’列彈出。ID = train_data.pop('id')# print train_data[0:1]# 將train_data中的‘species’列彈出。y = train_data.pop('species')# 將species向量化。y = LabelEncoder().fit(y).transform(y)print y# standardize the data by setting the mean to 0 and std to 1standardize = TrueX = StandardScaler().fit(train_data).transform(train_data) if standardize else train_data.valuesprint X[0:1]
2）pandas的主要數(shù)據(jù)類型dtype：object, 字符串類型int, 整型float, 浮點(diǎn)型datetime, 時(shí)間類型bool, 布爾型print data.dtpyes輸出每一列的數(shù)據(jù)類型3）索引4）選擇數(shù)據(jù)data.loc[1] //返回單列數(shù)據(jù)data.loc[1:3] //返回切片列數(shù)據(jù)，相當(dāng)于 data.loc[[1,2,3]]data.loc[:,['ID’]]data.loc[:4,['ID','YELP’]] //返回指定行的指定類，從0行到4行，包括第4行data.iloc[:,:] //返回所有數(shù)據(jù)data.iloc[:2,1:3] //返回特定行特定列的數(shù)據(jù)data[‘ID’] //返回列data[2:5] //返回行data[3:6][:2]data[‘ID’][3:6]data[3:6][‘ID’]data[data.YELP>0] //YELP這列的值不為空，即NaNdata[data['ID'].isin(['v4','v5'])] //返回有這個(gè)值的列5）缺失值處理去掉包含缺失值的行：df.dropna(how=‘a(chǎn)ny’)對缺失值進(jìn)行填充：df.fillna(values=‘NULL’)對數(shù)據(jù)進(jìn)行布爾補(bǔ)充：pandas.isnull(df)
6）數(shù)據(jù)處理
<class 'pandas.core.series.Series'>方法to_stringto_jsonjson.loads(df.loc[0:5,['ID','YELP']].to_json())輸出.csv文件。對應(yīng)解析1:
output = model.predict_proba(test_data.values[0::, 1::])        out = pd.DataFrame(output, index=test_data.values[0::, 0].astype(np.int), columns = species.index.values)out.index.name = "id"out.to_csv("output.csv")對應(yīng)解析2:
LABELS = sorted(pd.read_csv('train.csv').species.unique())ID, test_data = load_test_data()yPred_proba = model.predict(test_data)# Converting the test predictions in a dataframe as depicted by sample submissionyPred = pd.DataFrame(yPred_proba,index=ID,columns=LABELS)print('Creating and writing submission...')fp = open('submit.csv', 'w')fp.write(yPred.to_csv())print('Finished writing submission')# Display the submissionyPred.tail()