This repository has been archived by the owner on Jul 20, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 39
/
zeroconf.py
256 lines (217 loc) · 11 KB
/
zeroconf.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
# -*- coding: utf-8 -*-
# vim: tabstop=4 expandtab shiftwidth=4 softtabstop=4
"""
Copyright 2017 PayPal
Created on Mon Feb 27 19:11:59 PST 2017
@author: ekobylkin
@version 0.2
@author: ulrich arndt - data2knowledge
@update: 2017-09-27
"""
import argparse
import numpy as np
import os
import pandas as pd
import shutil
from sklearn.model_selection import train_test_split
from sklearn.metrics import (confusion_matrix, precision_score,
recall_score, f1_score, accuracy_score)
import utility as utl
import dataTransformationProcessing as dt
parameter = utl.init_process(__file__)
###########################################################
# define the command line argument parser
###########################################################
# https://docs.python.org/2/howto/argparse.html
parser = argparse.ArgumentParser(
description='zero configuration predictic modeling script. Requires a pandas HDFS dataframe file ' + \
'and a ymal parameter file as input as input')
parser.add_argument('-d',
'--data_file',
nargs=1,
help='input pandas HDFS dataframe .h5 with an unique indentifier and a target column\n' +
'as well as additional data columns\n'
'default values are cust_id and category or need to be defined in an\n' +
'optional parameter file '
)
parser.add_argument('-p',
'--param_file',
help='input yaml parameter file'
)
args = parser.parse_args()
logger = utl.get_logger(os.path.basename(__file__))
logger.info("Program started with the following arguments:")
logger.info(args)
###########################################################
# set dir to project dir
###########################################################
abspath = os.path.abspath(__file__)
dname = os.path.dirname(os.path.dirname(abspath))
os.chdir(dname)
###########################################################
# file check for the parameter
###########################################################
param_file = ''
if args.param_file:
param_file = args.param_file[0]
else:
param_file = os.path.abspath("./parameter/default.yml")
logger.info("Using the default parameter file: " + param_file)
if (not (os.path.isfile(param_file))):
msg = 'the input parameter file: ' + param_file + ' does not exist!'
logger.error(msg)
exit(8)
data_file = ''
if args.data_file:
data_file = args.data_file[0]
else:
msg = "A data file is mandatory!"
logger.error(msg)
exit(8)
if (not (os.path.isfile(data_file))):
msg = 'the input parameter file: ' + data_file + ' does not exist!'
logger.error(msg)
exit(8)
parameter = utl.read_parameter(param_file, parameter)
parameter["data_file"] = os.path.abspath(data_file)
parameter["basedir"] = os.path.abspath(parameter["basedir"])
parameter["parameter_file"] = os.path.abspath(param_file)
parameter["resultfile"] = os.path.abspath(parameter["resultfile"])
###########################################################
# set base dir
###########################################################
os.chdir(parameter["basedir"])
logger.info("Set basedir to: " + parameter["basedir"])
logger = utl.get_logger(os.path.basename(__file__))
logger.info("Program Call Parameter (Arguments and Parameter File Values):")
for key in sorted(parameter.keys()):
logger.info(" " + key + ": " + str(parameter[key]))
work_dir = parameter["workdir"]
result_filename = parameter["resultfile"]
atsklrn_tempdir = os.path.join(work_dir, 'atsklrn_tmp')
shutil.rmtree(atsklrn_tempdir, ignore_errors=True) # cleanup - remove temp directory
# if the memory limit is lower the model can fail and the whole process will crash
memory_limit = parameter["memory_limit"] # MB
global max_classifier_time_budget
max_classifier_time_budget = parameter["max_classifier_time_budget"] # but 10 minutes is usually more than enough
max_sample_size = parameter["max_sample_size"] # so that the classifiers fit method completes in a reasonable time
dataframe = dt.read_dataframe_h5(data_file, logger)
logger.info("Values of y " + str(dataframe[parameter["target_field"]].unique()))
logger.info("We need to protect NAs in y from the prediction dataset so we convert them to -1")
dataframe[parameter["target_field"]] = dataframe[parameter["target_field"]].fillna(-1)
logger.info("New values of y " + str(dataframe[parameter["target_field"]].unique()))
logger.info("Filling missing values in X with the most frequent values")
dataframe = dataframe.fillna(dataframe.mode().iloc[0])
logger.info("Factorizing the X")
# we need this list of original dtypes for the Autosklearn fit, create it before categorisation or split
col_dtype_dict = {col: ('Numerical' if np.issubdtype(dataframe[col].dtype, np.number) else 'Categorical')
for col in dataframe.columns if col not in [parameter["id_field"], parameter["target_field"]]}
# http://stackoverflow.com/questions/25530504/encoding-column-labels-in-pandas-for-machine-learning
# http://stackoverflow.com/questions/24458645/label-encoding-across-multiple-columns-in-scikit-learn?rq=1
# https://github.com/automl/auto-sklearn/issues/121#issuecomment-251459036
for col in dataframe.select_dtypes(exclude=[np.number]).columns:
if col not in [parameter["id_field"], parameter["target_field"]]:
dataframe[col] = dataframe[col].astype('category').cat.codes
df_unknown = dataframe[dataframe[parameter["target_field"]] == -1] # 'None' gets categorzized into -1
df_known = dataframe[dataframe[parameter["target_field"]] != -1] # not [0,1] for multiclass labeling compartibility
logger.debug("Length of unknown dataframe:" + str(len(df_unknown)))
logger.debug("Length of known dataframe:" + str(len(df_known)))
del dataframe
X, y = dt.x_y_dataframe_split(df_known, parameter)
logger.info("Preparing a sample to measure approx classifier run time and select features")
dataset_size = df_known.shape[0]
if dataset_size > max_sample_size:
sample_factor = dataset_size / float(max_sample_size)
logger.info("Sample factor =" + str(sample_factor))
X_sample, y_sample = dt.x_y_dataframe_split(df_known.sample(max_sample_size, random_state=42), parameter)
X_train, X_test, y_train, y_test = dt.train_test_split(X.copy(), y, stratify=y, test_size=33000,
random_state=42) # no need for larger test
else:
sample_factor = 1
X_sample, y_sample = X.copy(), y
X_train, X_test, y_train, y_test = train_test_split(X.copy(), y, stratify=y, test_size=0.33, random_state=42)
logger.info("train size:" + str(len(X_train)))
logger.info("test size:" + str(len(X_test)))
logger.info("Reserved 33% of the training dataset for validation (upto 33k rows)")
per_run_time_limit = dt.max_estimators_fit_duration(X_train.values, y_train, max_classifier_time_budget, logger,
sample_factor)
logger.info("per_run_time_limit=" + str(per_run_time_limit))
pool_size = dt.define_pool_size(int(memory_limit))
logger.info("Process pool size=" + str(pool_size))
feat_type = [col_dtype_dict[col] for col in X.columns]
logger.info("Starting autosklearn classifiers fiting on a 67% sample up to 67k rows")
dt.train_multicore(X_train.values, y_train, feat_type, int(memory_limit), atsklrn_tempdir, pool_size,
per_run_time_limit)
ensemble = dt.zeroconf_fit_ensemble(y_train, atsklrn_tempdir)
logger = utl.get_logger(os.path.basename(__file__))
logger.info("Validating")
logger.info("Predicting on validation set")
y_hat = ensemble.predict(X_test.values)
logger.info("#" * 72)
logger.info("Accuracy score {0:2.0%}".format(accuracy_score(y_test, y_hat)))
logger.info("The below scores are calculated for predicting '1' category value")
logger.info("Precision: {0:2.0%}, Recall: {1:2.0%}, F1: {2:.2f}".format(
precision_score(y_test, y_hat), recall_score(y_test, y_hat), f1_score(y_test, y_hat)))
#############################
## Print COnfusion Matrix
#############################
logger.info("Confusion Matrix: https://en.wikipedia.org/wiki/Precision_and_recall")
cm = confusion_matrix(y_test, y_hat)
for row in cm:
logger.info(row)
baseline_1 = str(sum(a for a in y_test))
baseline_all = str(len(y_test))
baseline_prcnt = "{0:2.0%}".format(float(sum(a for a in y_test) / len(y_test)))
logger.info("Baseline %s positives from %s overall = %1.1f%%" %
(sum(a for a in y_test), len(y_test), 100 * sum(a for a in y_test) / len(y_test)))
logger.info("#" * 72)
if df_unknown.shape[0] == 0: # if there is nothing to predict we can stop already
logger.info("##### Nothing to predict. Prediction dataset is empty. #####")
exit(0)
X_unknown, y_unknown, row_id_unknown = dt.x_y_dataframe_split(df_unknown, parameter, id=True)
logger.info("Re-fitting the model ensemble on full known dataset to prepare for prediciton. This can take a long time.")
try:
ensemble.refit(X.copy().values, y)
except Exception as e:
logger.info("Refit failed, reshuffling the rows, restarting")
logger.info(e)
try:
X2 = X.copy().values
indices = np.arange(X2.shape[0])
np.random.shuffle(indices) # a workaround to algoritm shortcomings
X2 = X2[indices]
y = y[indices]
ensemble.refit(X2, y)
except Exception as e:
logger.info("Second refit failed")
logger.info(e)
logger.info(
" WORKAROUND: because Refitting fails due to an upstream bug https://github.com/automl/auto-sklearn/issues/263")
logger.info(" WORKAROUND: we are fitting autosklearn classifiers a second time, now on the full dataset")
dt.train_multicore(X.values, y, feat_type, int(memory_limit), atsklrn_tempdir, pool_size, per_run_time_limit)
ensemble = dt.zeroconf_fit_ensemble(y_train, atsklrn_tempdir)
logger.info("Predicting. This can take a long time for a large prediction set.")
try:
y_pred = ensemble.predict(X_unknown.copy().values)
logger.info("Prediction done")
except Exception as e:
logger.info(e)
logger.info(
" WORKAROUND: because REfitting fails due to an upstream bug https://github.com/automl/auto-sklearn/issues/263")
logger.info(" WORKAROUND: we are fitting autosklearn classifiers a second time, now on the full dataset")
dt.train_multicore(X.values, y, feat_type, int(memory_limit), atsklrn_tempdir, pool_size, per_run_time_limit)
ensemble = dt.zeroconf_fit_ensemble(y_train, atsklrn_tempdir)
logger.info("Predicting. This can take a long time for a large prediction set.")
try:
y_pred = ensemble.predict(X_unknown.copy().values)
logger.info("Prediction done")
except Exception as e:
logger.info("##### Prediction failed, exiting! #####")
logger.info(e)
exit(2)
result_df = pd.DataFrame(
{parameter["id_field"]: row_id_unknown, 'prediction': pd.Series(y_pred, index=row_id_unknown.index)})
logger.info("Exporting the data")
result_df.to_csv(result_filename, index=False, header=True)
logger.info("##### Zeroconf Script Completed! #####")
utl.end_proc_success(parameter, logger)