-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdata_transform.py
375 lines (321 loc) · 18.8 KB
/
data_transform.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Sat Aug 19 15:11:25 2017
@author: ldong
"""
import numpy as np
import cPickle as pk
import pandas as pd
import matplotlib.pyplot as plt
import datetime as dt
import os
from sklearn import preprocessing
from geopy.distance import vincenty
from sklearn import neighbors
#from fancyimpute import KNN #SoftImpute, IterativeSVD, BiScaler
dir_path = os.path.dirname(os.path.realpath(__file__))
os.chdir(dir_path)
#train_y = pd.read_csv('train_2016_v2.csv', low_memory=False).set_index('parcelid').sort_values('transactiondate')
train_y = pd.read_csv('../../input/train_2016_v2.csv', low_memory=False)#.sort_values('transactiondate')
#train_y = train_y[~train_y.index.duplicated(keep='first')] # remove multiple solds, keep the latest sold
#train_y.sort_values('parcelid', inplace=True)
all_x = pd.read_csv('../../input/properties_2016.csv', low_memory=False)
#pred_date = ['2016-10-01' for _ in xrange(all_x.shape[0])]
#all_x = pd.concat([pd.DataFrame({'transactiondate':pred_date},index=all_x.index), all_x], axis=1)
train_y_17 = pd.read_csv('../../input/train_2017.csv', low_memory=False)#.sort_values('parcelid')
all_x_17 = pd.read_csv('../../input/properties_2017.csv', low_memory=False)
def property_transform(data_x,pic_name):
data_x.rename(columns = {#'transactiondate':'date_trans',
'yearbuilt':'date_build_year',
'basementsqft':'area_basement',
'yardbuildingsqft17':'area_patio',
'yardbuildingsqft26':'area_shed',
'poolsizesum':'area_pool',
'lotsizesquarefeet':'area_lot',
'garagetotalsqft':'area_garage',
'finishedfloor1squarefeet':'area_firstfloor_finished',
'calculatedfinishedsquarefeet':'area_total_calc',
'finishedsquarefeet6':'area_base',
'finishedsquarefeet12':'area_live_finished',
'finishedsquarefeet13':'area_liveperi_finished',
'finishedsquarefeet15':'area_total_finished',
'unitcnt':'num_unit',
'numberofstories':'num_story',
'roomcnt':'num_room',
'calculatedbathnbr':'num_50_bath',
'bedroomcnt':'num_bedroom',
'fullbathcnt':'num_100_bath',
'threequarterbathnbr':'num_75_bath',
'fireplacecnt':'num_fireplace',
'poolcnt':'num_pool',
'garagecarcnt':'num_garage',
'regionidcounty':'categ_geo_region_cnty',
'regionidcity':'categ_geo_region_city',
'regionidzip':'categ_geo_region_zip',
'regionidneighborhood':'categ_geo_region_neighbor',
'latitude':'geo_lati',
'longitude':'geo_longi',
'censustractandblock':'geo_census_cnty_trac_bloc',
'rawcensustractandblock':'geo_census_cnty_trac_bloc_raw',
'taxvaluedollarcnt':'val_total',
'structuretaxvaluedollarcnt':'val_building',
'landtaxvaluedollarcnt':'val_land',
'taxamount':'val_tax',
'assessmentyear':'date_tax_year',
'taxdelinquencyflag':'flag_tax_delinquency',
'taxdelinquencyyear':'date_tax_delinquency_year',
'propertyzoningdesc':'categ_zone_desc',
'propertylandusetypeid':'categ_zone_type',
'propertycountylandusecode':'categ_zone_cnty',
'buildingqualitytypeid':'categ_quality',
'buildingclasstypeid':'categ_framing',
'typeconstructiontypeid':'categ_material',
'decktypeid':'categ_deck',
'storytypeid':'categ_story',
'heatingorsystemtypeid':'categ_heat',
'airconditioningtypeid':'categ_aircon',
'architecturalstyletypeid':'categ_architec',
'hashottuborspa':'flag_tub_spa',
'pooltypeid2':'flag_pool_w_spa_tub',
'pooltypeid7':'flag_pool_wo_spa_tub'}, inplace=True)
data_x.drop('finishedsquarefeet50', axis=1, inplace=True) # repeat with finishedfloor1squarefeet
data_x.drop('bathroomcnt', axis=1, inplace=True) # repeat with bathroomcnt
data_x.drop('pooltypeid10', axis=1, inplace=True) # repeat with hashottuborspa
data_x.drop('fireplaceflag', axis=1, inplace=True) # redundant due to fireplacecnt
data_x.drop('num_room', axis=1, inplace=True) # != bath + bed ???
data_x.drop('date_tax_year', axis=1, inplace=True) # all 2015
data_x['num_pool'].fillna(value=0, inplace=True)
data_x['num_fireplace'].fillna(value=0, inplace=True)
data_x['num_75_bath'].fillna(value=0, inplace=True)
data_x['flag_tub_spa'].fillna(value=0, inplace=True)
data_x['flag_pool_w_spa_tub'].fillna(value=0, inplace=True)
data_x['flag_pool_wo_spa_tub'].fillna(value=0, inplace=True)
data_x['flag_tax_delinquency'].fillna(value=0, inplace=True)
data_x['date_build_year'] = 2018. - data_x['date_build_year']
data_x.rename(columns={'date_build_year':'age'}, inplace=True)
data_x['flag_tub_spa'] = data_x['flag_tub_spa'].values.astype(float)
data_x['num_50_bath'] = (data_x['num_50_bath'] - data_x['num_100_bath'])/0.5
data_x.loc[data_x['flag_tax_delinquency']=='Y', 'flag_tax_delinquency'] = 1
data_x['geo_lati'] = data_x['geo_lati']#/1000000.0
data_x['geo_longi'] = data_x['geo_longi']#/1000000.0
data_x['categ_zone_desc'] = data_x['categ_zone_desc'].astype('category').cat.codes
data_x['categ_zone_type'] = data_x['categ_zone_type'].astype('category').cat.codes
data_x['categ_zone_cnty'] = data_x['categ_zone_cnty'].astype('category').cat.codes
data_x['categ_aircon'] = data_x['categ_aircon'].astype('category').cat.codes
data_x['categ_quality'] = data_x['categ_quality'].astype('category').cat.codes
data_x['categ_framing'] = data_x['categ_framing'].astype('category').cat.codes
data_x['categ_story'] = data_x['categ_story'].astype('category').cat.codes
data_x['categ_architec'] = data_x['categ_architec'].astype('category').cat.codes
data_x['categ_deck'] = data_x['categ_deck'].astype('category').cat.codes
data_x['categ_heat'] = data_x['categ_heat'].astype('category').cat.codes
data_x['categ_material'] = data_x['categ_material'].astype('category').cat.codes
data_x['categ_geo_region_cnty'] = data_x['categ_geo_region_cnty'].astype('category').cat.codes
data_x['categ_geo_region_city'] = data_x['categ_geo_region_city'].astype('category').cat.codes
data_x['categ_geo_region_zip'] = data_x['categ_geo_region_zip'].astype('category').cat.codes
data_x['categ_geo_region_neighbor'] = data_x['categ_geo_region_neighbor'].astype('category').cat.codes
data_x.fillna(value=-1,inplace=True)
census_cnty_trac_bloc_raw_tmp = data_x.loc[:,'geo_census_cnty_trac_bloc_raw']
census_cnty_trac_bloc_tmp = data_x.loc[:,'geo_census_cnty_trac_bloc']
census_cnty_trac_bloc_tmp[census_cnty_trac_bloc_tmp==-1] = census_cnty_trac_bloc_raw_tmp[census_cnty_trac_bloc_tmp==-1]*1000000
census_cnty_trac_bloc_tmp[census_cnty_trac_bloc_tmp==-1000000] = -1
census_cnty_trac_bloc_tmp = census_cnty_trac_bloc_tmp.apply(long).apply(str)
data_x['categ_geo_census_cnty'] = pd.Series(census_cnty_trac_bloc_tmp.str[0:4].astype(float), index=data_x.index)
census_trac_tmp = census_cnty_trac_bloc_tmp.str[4:10]
census_trac_tmp[census_trac_tmp=='']=-1
data_x['categ_geo_census_trac'] = pd.Series(census_trac_tmp.astype(float), index=data_x.index)
census_bloc_tmp = census_cnty_trac_bloc_tmp.str[10:14]
census_bloc_tmp[census_bloc_tmp=='']=-1
data_x['categ_geo_census_bloc'] = pd.Series(census_bloc_tmp.astype(float), index=data_x.index)
data_x['categ_geo_census_cnty_trac'] = pd.Series(census_cnty_trac_bloc_tmp.str[0:10].astype(float), index=data_x.index)
data_x.drop('geo_census_cnty_trac_bloc_raw', axis=1, inplace=True) # sort-of repeat with censustractandblock
data_x.drop('fips', axis=1, inplace=True) # repeat with census_cnty
data_x.drop('categ_geo_census_cnty', axis=1, inplace=True) # repeat with categ_georegion_cnty
data_x.drop('categ_geo_census_trac', axis=1, inplace=True)
data_x.drop('categ_geo_census_bloc', axis=1, inplace=True)
data_x.drop('geo_census_cnty_trac_bloc', axis=1, inplace=True)
data_x['categ_geo_census_cnty_trac'] = data_x['categ_geo_census_cnty_trac'].astype('category').cat.codes
data_x['categ_geo_census_cnty_trac'] = data_x['categ_geo_census_cnty_trac'] - 1. # cuz fillna before
# data_x['date_trans'] = data_x['date_trans'].astype('datetime64[ns]')
# #data_x = pd.concat([pd.DataFrame({'date':all_x['data_trans'].dt.year*100+all_x['date_trans'].dt.month}), data_x], axis=1)
# data_x = pd.concat([pd.DataFrame({'date': np.ceil(data_x['date_trans'].dt.month.astype(float)/3.) + 4.*(np.ceil(all_x['date_trans'].dt.year.astype(float)/2016.)-1.) }), data_x], axis=1) # use quarter
# data_x.drop('date_trans', axis=1, inplace=True)
na_pct = (data_x==-1).sum(axis=0) / float(data_x.shape[0])
plt.rcParams['figure.figsize'] = 7, 15
plt.figure()
ax = na_pct.sort_values().plot(kind='barh')
ax.set_xlabel('missing_value_penctage')
ax.get_figure().savefig('missing_value_pct'+pic_name)
return data_x, na_pct
all_x, na_pct_all = property_transform(all_x, '_all')
all_x_17, na_pct_all_17 = property_transform(all_x_17, '_all_17')
## include 16 tax values in 17
#all_x_17.rename(columns={'val_tax':'val_tax_17',
# 'val_total':'val_total_17',
# 'val_building':'val_building_17',
# 'val_land':'val_land_17'},inplace=True)
#
#all_x_17 = all_x_17.merge(all_x[['parcelid','val_tax','val_total','val_building','val_land']], on='parcelid', how='left')
#
#all_x_17.rename(columns={'val_tax':'val_tax_16',
# 'val_total':'val_total_16',
# 'val_building':'val_building_16',
# 'val_land':'val_land_16'},inplace=True)
#
#all_x_17.rename(columns={'val_tax_17':'val_tax',
# 'val_total_17':'val_total',
# 'val_building_17':'val_building',
# 'val_land_17':'val_land'},inplace=True)
#flag_na_pct = na_pct_all < 0.83 #threshold_na
#all_x = all_x[all_x.columns[flag_na_pct]]
#all_x_17 = all_x_17[all_x_17.columns[flag_na_pct]]
#with open('prop_2016_clean.pkl','wb') as f:
# pk.dump(all_x, f, protocol=pk.HIGHEST_PROTOCOL)
def create_train(train_y, all_x, pic_name):
train_x = train_y.merge(all_x, on='parcelid', how='left')
train_x.rename(columns={'transactiondate':'date'}, inplace=True)
train_x['date'] = train_x['date'].astype('datetime64[ns]')
plt.rcParams['figure.figsize'] = 7, 7
plt.figure()
ax = train_x['date'].groupby([train_x["date"].dt.year, train_x["date"].dt.month]).count().plot(kind="bar")
ax.get_figure().savefig('train_date'+pic_name)
# train_x = pd.concat([pd.DataFrame({'date_trans': np.ceil(train_x['date'].dt.month.astype(float)/3.) + 4.*(np.ceil(train_x['date'].dt.year.astype(float)/2016.)-1.) }), train_x], axis=1) # use quarter
train_x = pd.concat([pd.DataFrame({'date_trans': np.ceil(train_x['date'].dt.month) + 12.*(np.ceil(train_x['date'].dt.year.astype(float)/2016.)-1.) }), train_x], axis=1) # use quarter
train_x.drop('date', axis=1, inplace=True)
train_x.rename(columns={'date_trans':'date'}, inplace=True)
na_pct_train = (train_x==-1).sum(axis=0) / float(train_x.shape[0])
train_x.drop('logerror', axis=1, inplace=True)
train_y.drop('transactiondate', axis=1, inplace=True)
# train_y.drop('parcelid', axis=1, inplace=True)
## remove outliers
fac_std = 5.2
threshold_logerror = [float(i) for i in [train_y.logerror.mean(axis=0)-fac_std*train_y.logerror.std(axis=0), train_y.logerror.mean(axis=0)+fac_std*train_y.logerror.std(axis=0)] ]
flag_logerror = (train_y.logerror>threshold_logerror[0]) & (train_y.logerror<threshold_logerror[1])
# threshold_logerror = [-0.4, 0.418 ]
flag_logerror = (train_y.logerror>threshold_logerror[0]) & (train_y.logerror<threshold_logerror[1])
train_x = train_x[flag_logerror]
train_y = train_y[flag_logerror]
train_x.reset_index(drop=True,inplace=True)
train_y.reset_index(drop=True,inplace=True)
# limit outlier values
# train_y.loc[:,'logerror'] = train_y.logerror.clip(-0.08,0.08)
return train_x, train_y, na_pct_train
train_x, train_y, na_pct_train = create_train(train_y, all_x, '_16')
train_x_17, train_y_17, na_pct_train_17 = create_train(train_y_17, all_x_17, '_17')
#all_x.drop('parcelid', axis=1, inplace=True)
#all_x_17.drop('parcelid', axis=1, inplace=True)
#train_x.drop('parcelid', axis=1, inplace=True)
#train_x_17.drop('parcelid', axis=1, inplace=True)
flag_all_nan = all_x.geo_lati == -1 # flag for properties with no features at all
flag_all_nan_17 = all_x_17.geo_lati == -1 # flag for properties with no features at all
#def rows_filter(flag_train, data_x, train_y, fac_std):
# print 'original num of rows = ', data_x.shape[0], ' columns = ', data_x.shape[1]
#
# threshold_logerror = [float(i) for i in [train_y.mean(axis=0)-fac_std*train_y.std(axis=0), train_y.mean(axis=0)+fac_std*train_y.std(axis=0)] ]
# flag_logerror = (train_y.logerror>threshold_logerror[0]) & (train_y.logerror<threshold_logerror[1])
#
## flag_date_valid = ( (data_x['date'].dt.month == 10) & (data_x['date'].dt.day > 14) ) | ( data_x['date'].dt.month > 10 )
# flag_date_valid = np.squeeze(np.zeros([data_x.shape[0],1], dtype='bool')) # no validation data, use cross validation
#
# valid_x = data_x[flag_logerror & flag_date_valid]
# data_x = data_x[flag_logerror & (~ flag_date_valid)]
#
# valid_y = train_y[flag_logerror & flag_date_valid]
# train_y = train_y[flag_logerror & (~ flag_date_valid)]
#
# return valid_x, valid_y, data_x, train_y
#
#valid_x, valid_y, train_x, train_y = rows_filter(1, train_x, train_y, 5.2)
#valid_x.drop('parcelid', axis=1, inplace=True)
# make sure the parcelid order
sub = pd.read_csv('../../input/sample_submission.csv')
sub.rename(columns={'ParcelId':'parcelid'}, inplace=True)
sub.drop(['201610','201611','201612','201710','201711','201712'],axis=1,inplace=True)
all_x = sub.merge(all_x,how='left',on='parcelid')
all_x_17 = sub.merge(all_x_17,how='left',on='parcelid')
#sub_pid = list(sub.parcelid)
#all_x = all_x.set_index('parcelid')
#all_x = all_x.reindex(sub_pid)
#all_x['parcelid'] = all_x.index
#all_x = all_x[-1:] + all_x[:-1]
#
#all_x_17 = all_x_17.set_index('parcelid')
#all_x_17 = all_x_17.reindex(sub_pid)
#all_x_17['parcelid'] = all_x_17.index
#all_x_17 = all_x_17[-1:] + all_x_17[:-1]
# !!! remember to take care of those rows with no features!!!
with open('bundle_woImpute.pkl', 'wb') as f:
pk.dump([train_x, train_y, all_x, flag_all_nan, na_pct_all, train_x_17, train_y_17, all_x_17, flag_all_nan_17], f, protocol=pk.HIGHEST_PROTOCOL)
### imputation
### simple one
##fill_NaN = preprocessing.Imputer(missing_values=-1, strategy='median', axis=0)
##all_x = pd.DataFrame(fill_NaN.fit_transform(all_x), columns=all_x.columns, index=all_x.index)
## complex ones
##all_x.replace(to_replace=-1,value=np.nan, inplace=True)
##all_x = pd.DataFrame(SoftImpute().complete(all_x), columns=all_x.columns, index=all_x.index)
##all_x = pd.DataFrame(IterativeSVD().complete(all_x), columns=all_x.columns, index=all_x.index)
#
#def knn_impute(x, y_train, flagcateg):
# flagnan = y_train.isnull().values
# x_train = x[np.invert(flagnan)]
# x_train_mean = x_train.mean()
# x_train_max = x_train.max()
# x_train_min = x_train.min()
# x_train = (x_train - x_train_mean) / (x_train_max - x_train_min)
# y_train = y_train[np.invert(flagnan)]
# x_test = x[flagnan]
# x_train = x_train.values.astype(np.float32, copy=False)
# y_train = y_train.values.astype(np.float32, copy=False)
# x_test = x_test.values.astype(np.float32, copy=False)
# n_neighbors = int(np.sqrt(x_train.shape[0]))
# if flagcateg:
# knn = neighbors.KNeighborsClassifier(n_neighbors, weights='distance', n_jobs=8)
# else:
# knn = neighbors.KNeighborsRegressor(n_neighbors, weights='distance', n_jobs=8)
# y_test = knn.fit(x_train, y_train).predict(x_test)
# return y_test
## return pd.DataFrame(y_test, columns=y_data[flagIncomp].columns, index=y_data[flagIncomp].index)
#
#def impute(na_pct_train, train_x, train_x_17, train_y, train_y_17, flag_train, flag_16):
# na_pct_train.drop(['parcelid'], axis=0, inplace=True)
#
# if flag_train:
# if flag_16:
# all_data = train_x.join(train_y, how='left')
# else:
# all_data = train_x.append(train_x_17, ignore_index=True)
# else:
# if flag_16:
# all_data = train_x
# else:
# all_data = train_x_17
#
# all_data.replace(to_replace=-1,value=np.nan, inplace=True)
# flagIncomp = list(na_pct_train[na_pct_train>0].index)
# flagComp = list(na_pct_train[na_pct_train==0].index)
#
# for i in xrange(len(flagIncomp)):
# print flagIncomp[i]
# y_train = all_data[flagIncomp[i]]
# flagnan = y_train.isnull().values
# if ('categ' in flagIncomp[i]) or ('num' in flagIncomp[i]) or ('flag' in flagIncomp[i]):
# train_x[flagIncomp[i]].values[flagnan] = knn_impute(all_data[flagComp], y_train, 1)
# else:
# train_x[flagIncomp[i]].values[flagnan] = knn_impute(all_data[flagComp], y_train, 0)
# print 'Oh yeah'
# return train_x
#
#train_x = impute(na_pct_train, train_x, train_x_17, train_y, train_y_17, 1, 1)
#train_x_17 = impute(na_pct_train_17, train_x, train_x_17, train_y, train_y_17, 1, 0)
#all_x.values[np.invert(flag_all_nan)] = impute(na_pct_train, all_x[np.invert(flag_all_nan)], all_x_17[np.invert(flag_all_nan_17)], train_y, train_y_17, 0, 1)
#all_x_17.values[np.invert(flag_all_nan_17)] = impute(na_pct_train_17, all_x[np.invert(flag_all_nan)], all_x_17[np.invert(flag_all_nan_17)], train_y, train_y_17, 0, 0)
#
#
##all_x.replace(to_replace=-1, value=np.nan, inplace=True)
##all_x_mean = all_x.mean()
##all_x_max = all_x.max()
##all_x_min = all_x.min()
##all_x = (all_x - all_x_mean) / (all_x_max - all_x_min)
#
##all_x = pd.DataFrame(KNN(k=int(np.sqrt(all_x.shape[0]))).complete(all_x), columns=all_x.columns, index=all_x.index)
##all_x = all_x * (all_x_max - all_x_min) + all_x_mean