#coding:utf8import pandas as pdimport numpy as npfrom pandas import Series,DataFrame # 如果有id列，则需先删除id列再进行对应操作，最后再补上# 统计的时候不需要用到id列,删除的时候需要考虑# delete rowdef row_del(df, num_percent, label_len = 0): #print list(df.count(axis=1)) col_num = len(list(list(df.values)[1])) - label_len # -1为考虑带标签 if col_num<0: print 'Error' #print int(col_num*num_percent) return df.dropna(axis=0, how='any', thresh=int(col_num*num_percent)) # 如果有字符串类型，则报错# data normalization -1 to 1# label_col: 不需考虑的类标，可以为字符串或字符串列表# 数值类型统一到float64def data_normalization(df, label_col = []): lab_len = len(label_col) print label_col if lab_len>0: df_temp = df.drop(label_col, axis = 1) df_lab = df[label_col] print df_lab else: df_temp = df max_val = list(df_temp.max(axis=0)) min_val = list(df_temp.min(axis=0)) mean_val = list((df_temp.max(axis=0) + df_temp.min(axis=0)) / 2) nan_values = df_temp.isnull().values row_num = len(list(df_temp.values)) col_num = len(list(df_temp.values)[1]) for rn in range(row_num): #data_values_r = list(data_values[rn]) nan_values_r = list(nan_values[rn]) for cn in range(col_num): if nan_values_r[cn] == False: df_temp.values[rn][cn] = 2 * (df_temp.values[rn][cn] - mean_val[cn])/(max_val[cn] - min_val[cn]) else: print 'Wrong' for index,lab in enumerate(label_col): df_temp.insert(index, lab, df_lab[lab]) return df_temp # 创建一个带有缺失值的数据框：df = pd.DataFrame(np.random.randn(5,3), index=list('abcde'), columns=['one','two','three'])df.ix[1,:-1]=np.nandf.ix[1:-1,2]=np.nandf.ix[0,0]=int(1)df.ix[2,2]='abc' # 查看一下数据内容：print '/ndf1'print df print row_del(df, 0.8) print '-------------------------' df = data_normalization(df, ['two', 'three'])print df print df.dtypes print (type(df.ix[2,2])) 
