-
Notifications
You must be signed in to change notification settings - Fork 10
/
base.py
185 lines (170 loc) · 7.4 KB
/
base.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
"""
Module representing the Base-Learners, Members of an Ensemble
"""
import numpy as np
import os
from keras.models import load_model
class Member:
def __init__(self, name, train_probs, train_classes, val_probs,
val_classes, submission_probs):
"""
Constructor for an Ensemble Member (Base-Learner) based on class'
probabilities.
Args:
name: name of the member. Must be unique.
train_probs: class' probabilities for training the Meta-Learner
train_classes: ground truth (classes) for training the Meta-Learner
val_probs: class' probabilitiess for validating the Meta-Learner
val_classes: ground truth (classes) for validating the Meta-Learner
submission_probs: the final (submission) prediction probabilities
Returns:
a Member object
"""
self.name = name
self.train_probs = np.array(train_probs)
self.train_classes = np.array(train_classes)
self.val_probs = np.array(val_probs)
self.val_classes = np.array(val_classes)
self.submission_probs = np.array(submission_probs)
def __repr__(self):
return "<Member: " + self.name + ">"
def __str__(self):
return "Member: " + self.name
def __eq__(self, other):
c1 = self.name == other.name
c2 = np.array_equal(self.train_classes, other.train_classes)
c3 = np.array_equal(self.val_classes, other.val_classes)
c4 = np.array_equal(self.val_probs, other.val_probs)
c5 = np.array_equal(self.train_probs, other.train_probs)
return c1 and c2 and c3 and c4 and c5
@classmethod
def load(cls, folder=None):
"""
Loads base-learner from directory
Args:
folder: directory where member is saved
Returns:
loaded Member object
"""
name = folder.split(os.sep)[-1].replace(os.sep, "")
if folder[-1] == os.sep:
name = folder.split(os.sep)[-2].replace(os.sep, "")
train_probs = np.load(os.path.join(folder, "train_probs.npy"))
train_classes = np.load(os.path.join(folder, "train_classes.npy"))
val_probs = np.load(os.path.join(folder, "val_probs.npy"))
val_classes = np.load(os.path.join(folder, "val_classes.npy"))
submission_probs = None
if os.path.isfile(os.path.join(folder, "submission_probs.npy")):
submission_probs = np.load(
os.path.join(folder, "submission_probs.npy"))
member = Member(name, train_probs, train_classes, val_probs,
val_classes, submission_probs)
print("Loaded", name)
return member
def save(self, folder="./premodels/"):
"""
Saves member object to folder.
Args:
folder: the folder where models should be saved to.
Create if not exists.
"""
if folder[-1] != os.sep:
folder += os.sep
if not os.path.exists(folder):
os.mkdir(folder)
if not os.path.exists(folder + self.name):
os.mkdir(folder + self.name)
np.save(folder + self.name + "/val_probs.npy", self.val_probs)
np.save(folder + self.name + "/train_probs.npy", self.train_probs)
np.save(folder + self.name + "/val_classes.npy", self.val_classes)
np.save(folder + self.name + "/train_classes.npy", self.train_classes)
if self.submission_probs is not None:
np.save(folder + self.name + "/submission_probs.npy",
self.submission_probs)
class KerasMember(Member):
"""
Representation of a single keras model member (Base-Learner) of an Ensemble
"""
def __init__(self, name=None, keras_model=None, train_batches=None,
val_batches=None, submission_probs=None, keras_modelpath=None,
keras_kwargs={}):
"""
Constructor of a Keras Ensemble Member.
Internal class' probabilities are calculates based on DataGenerators.
Args:
name: name of the model. Must be unique.
model: the (pre-trained) keras model.
Or provide `keras_modelpath` instead.
train_batches: training data for the Meta-Learner.
Either a Keras `DataGenerator` or a tuple
with training set (X, y).
val_batches: validation data for the Meta-Learner.
Either a Keras `DataGenerator` or a tuple
with validation set (X, y).
submission_probs: the final submission prediction probabilities
keras_modelpath: path to load keras model from (if `model`
argument is None)
keras_kwargs: kwargs for keras `load_model` (if `model` argument
is None)
"""
assert(name is not None)
self.name = name
self.model = keras_model
self.submission_probs = submission_probs
# Initialize Params
self.val_probs = None
self.train_probs = None
self.val_classes = None
self.train_classes = None
if (keras_model is None) and (keras_modelpath is not None):
self.load_kerasmodel(self.keras_modelpath, self.keras_kwargs)
if train_batches is not None:
self._calculate_train_predictions(train_batches)
if val_batches is not None:
self._calculate_val_predictions(val_batches)
def _test_datatuple(self, datatuple):
assert(len(datatuple) == 2)
assert(datatuple[0].shape[0] == datatuple[1].shape[0])
def _calculate_predictions(self, batches):
if hasattr(batches, 'shuffle'):
batches.reset()
batches.shuffle = False
if type(batches) is tuple:
self._test_datatuple(batches)
return self.model.predict(batches[0])
return self.model.predict_generator(
batches, steps=(batches.n // batches.batch_size) + 1, verbose=1)
def _calculate_val_predictions(self, val_batches):
if type(val_batches) is tuple:
self.val_classes = val_batches[1]
elif hasattr(val_batches, 'classes'):
self.val_classes = np.array(val_batches.classes)
elif hasattr(val_batches, 'y'):
self.val_classes = np.array(val_batches.y)
else:
raise ValueError("No known class in data batch")
self.val_probs = self._calculate_predictions(val_batches)
return self.val_probs
def _calculate_train_predictions(self, train_batches):
if type(train_batches) is tuple:
self.train_classes = train_batches[1]
elif hasattr(train_batches, 'classes'):
self.train_classes = np.array(train_batches.classes)
elif hasattr(train_batches, 'y'):
self.train_classes = np.array(train_batches.y)
else:
raise ValueError("No known class in data batch")
self.train_probs = self._calculate_predictions(train_batches)
return self.train_probs
def load_kerasmodel(self, keras_modelpath=None, keras_kwargs={}):
"""
Utility method for loading Keras model
Args:
keras_modelpath: path to keras model
keras_kwargs: arguments for keras `load_model`
"""
if keras_kwargs is None:
keras_kwargs = {}
self.model = load_model(keras_modelpath, **keras_kwargs)
print("Keras Model Loaded:", keras_modelpath)
return self.model