-
Notifications
You must be signed in to change notification settings - Fork 0
/
util.py
223 lines (186 loc) · 6.96 KB
/
util.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import sys
import numpy as np
import pandas as pd
try:
from tensorflow.python.framework import dtypes
except ImportError:
pass
# modified from https://github.com/tensorflow/tensorflow/blob/7c36309c37b04843030664cdc64aca2bb7d6ecaa/tensorflow/contrib/learn/python/learn/datasets/mnist.py#L189
class Batcher(object):
def __init__(self,
features,
labels,
dtype=dtypes.float32):
"""Construct a DataSet.
"""
dtype = dtypes.as_dtype(dtype).base_dtype
if dtype not in (dtypes.uint8, dtypes.float32):
raise TypeError('Invalid image dtype %r, expected uint8 or float32' %
dtype)
assert features.shape[0] == labels.shape[0], (
'features.shape: %s labels.shape: %s' % (features.shape, labels.shape))
self._num_examples = features.shape[0]
# # Convert shape from [num examples, rows, columns, depth]
# # to [num examples, rows*columns] (assuming depth == 1)
# if reshape:
# assert features.shape[3] == 1
# features = features.reshape(features.shape[0],
# features.shape[1] * features.shape[2])
# if dtype == dtypes.float32:
# # Convert from [0, 255] -> [0.0, 1.0].
# features = features.astype(np.float32)
# features = np.multiply(features, 1.0 / 255.0)
self._features = features
self._labels = labels
self._epochs_completed = 0
self._index_in_epoch = 0
@property
def features(self):
return self._features
@property
def labels(self):
return self._labels
@property
def num_examples(self):
return self._num_examples
@property
def epochs_completed(self):
return self._epochs_completed
def next_batch(self, batch_size):
"""Return the next `batch_size` examples from this data set."""
# print('index({})/total({}), batch size: {}' \
# .format(self._index_in_epoch, self._num_examples, batch_size))
start = self._index_in_epoch
self._index_in_epoch += batch_size
if self._index_in_epoch > self._num_examples:
print('finished epoch {}'.format(self._epochs_completed))
# print('shuffle:')
self._epochs_completed += 1
# Shuffle the data
perm = np.arange(self._num_examples)
np.random.shuffle(perm)
self._features = self._features[perm]
self._labels = self._labels[perm]
# Start next epoch
start = 0
self._index_in_epoch = batch_size
assert batch_size <= self._num_examples
end = self._index_in_epoch
return self._features[start:end], self._labels[start:end]
class Cutter(object):
def __init__(self, method='cut', nbins=10, cutcols=None,
expand_left=-np.inf, expand_right=np.inf):
self._method = method
self._nbins = nbins
self._cutcols = cutcols
self._expand_left = expand_left
self._expand_right = expand_right
self._bins_list = []
if self._method == 'cut':
self._cut_func = pd.cut
elif self._method == 'qcut':
self._cut_func = pd.qcut
else:
raise NotImplementedError()
def fit(self, X, y=None):
self.fit_transform(X, y)
return self
def fit_transform(self, X, y=None):
if isinstance(X, (pd.DataFrame, pd.Series)):
X = X.values
self._ndim = len(X.shape)
if len(X.shape) == 1:
#self._n_features = 0
#val, bins = self.cut_1d(0, X, 0)
#self._levels = [val.categories.values]
#self._bins_list.append(bins)
#return val.codes
X = X.reshape((X.shape[0], 1))
elif len(X.shape) > 2:
raise NotImplementedError()
self._n_features = X.shape[1]
cutcols = self._get_cutcols()
ncols = len(cutcols)
cutted_list = [0] * ncols
self._bins_list = [0] * ncols
for i, icol in enumerate(cutcols):
val, bins = self._cut_1d(i, X[:, icol], ncols)
cutted_list[i] = val
self._bins_list[i] = bins
self._levels = [c.categories.values for c in cutted_list]
self._expand_levels_bins()
ret = np.column_stack([c.codes for c in cutted_list])
return ret
def _get_cutcols(self):
if self._cutcols is None:
cutcols = range(self._n_features)
else:
cutcols = self._cutcols
return cutcols
def _cut_1d(self, i, v, ncols):
if isinstance(self._nbins, (list, tuple)):
if len(self._nbins) != ncols:
raise ValueError('nbins must be scalar or list with same size' +
'to number of columns of input data')
nbins = self._nbins[i]
else:
nbins = self._nbins
val, bins = self._cut_func(v, nbins, retbins=True)
return val, bins
def _expand_levels_bins(self):
for lev, bins in zip(self._levels, self._bins_list):
if self._expand_left is not None:
lev[0] = pd.Interval(left=self._expand_left,
right=lev[0].right,
closed=lev[0].closed)
bins[0] = self._expand_left
if self._expand_right is not None:
lev[-1] = pd.Interval(left=lev[-1].left,
right=self._expand_right,
closed=lev[-1].closed)
bins[-1] = self._expand_right
def transform(self, X):
if isinstance(X, (pd.DataFrame, pd.Series)):
X = X.values
if self._n_features == 0:
if len(X.shape) != 1:
raise ValueError('X must be 1d vector')
return self._trans_1d(-1, X)
else:
if len(X.shape) != 2:
raise NotImplementedError()
cutcols = self._get_cutcols()
ncols = len(cutcols)
cutted_list = [0] * ncols
for i, icol in enumerate(cutcols):
val = self._trans_1d(i, X[:, icol])
cutted_list[i] = val
ret = np.column_stack([c for c in cutted_list])
return ret
def _trans_1d(self, i, v):
ret = pd.cut(v, bins=self._bins_list[i], labels=False, include_lowest=True)
return ret
def test_cutter():
import numpy.random as r
x = r.randn(10, 2)
cutter = Cutter(nbins=[3, 4])
print(x)
xc = cutter.fit_transform(x)
print(xc)
print('levels:', cutter._levels)
print(cutter._bins_list)
xc2 = cutter.transform(x)
print(xc2)
assert(np.all(xc == xc2))
from sklearn.preprocessing import OneHotEncoder
enc = OneHotEncoder()
dummy = enc.fit_transform(xc)
print('dummy:', dummy.toarray())
#print('dummy:', type(dummy))
def main():
return
if __name__ == '__main__':
#main()
test_cutter()