-
Notifications
You must be signed in to change notification settings - Fork 1
/
TrainModel.py
171 lines (134 loc) · 7.85 KB
/
TrainModel.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
from ProjectPaths import ProjectPaths
from PerformanceMetrics import PerformanceMetrics
from ModelFactory import ModelFactory
from Datasets import Datasets
from keras.callbacks import ModelCheckpoint, TensorBoard, EarlyStopping
from keras.optimizers import SGD, RMSprop
from keras.utils import to_categorical
from sklearn.metrics import confusion_matrix
from collections import namedtuple
import numpy as np
import pandas as pd
import csv
import sys
import os
RunSettings = namedtuple("RunSettings", ["model_name", "last_vgg_layer", "pre_trained_weights", "include_top", "all_trainable",
"dataset_name", "batch_size", "epochs", "optimizer", "lr",
"momentum", "decay", "nesterov", "one_hot"])
def run_name_for(settings):
return "{}_pr_{}_{}_al_{}_{}_e_{}_bs_{}_op_{}".format(settings.model_name, settings.last_vgg_layer, settings.pre_trained_weights,
settings.all_trainable, settings.dataset_name, settings.epochs,
settings.batch_size, settings.optimizer)
def optimizers():
return {"sgd": lambda settings: SGD(settings.lr, settings.momentum, settings.decay, settings.nesterov),
"rmsprop": lambda settings: RMSprop(settings.lr) if settings.lr else RMSprop()}
def optimizer_for(settings):
return optimizers()[settings.optimizer](settings)
def compile_model(model, settings):
model.compile(optimizer=optimizer_for(settings), loss='binary_crossentropy',
metrics=['accuracy', PerformanceMetrics.precision,
PerformanceMetrics.recall, PerformanceMetrics.fmeasure])
return model
def train_model(model, settings, train_images, train_labels, validation_images, validation_labels, verbose=True):
checkpoint_dir = ProjectPaths.checkpoint_dir_for(run_name_for(settings), settings.batch_size, settings.epochs)
if not os.path.exists(checkpoint_dir):
os.mkdir(checkpoint_dir)
file_in_checkpoint_dir = ProjectPaths.file_in_checkpoint_dir(run_name_for(settings), settings.batch_size,
settings.epochs, settings.model_name +
"__{epoch:02d}_{val_acc:.2f}.hdf5")
early_stopping_callback = EarlyStopping(patience=10)
model_checkpoint_callback = ModelCheckpoint(file_in_checkpoint_dir, monitor='val_acc', verbose=verbose,
save_weights_only=True,
save_best_only=True)
log_dir = os.path.join(ProjectPaths.log_dir(), run_name_for(settings))
tensorboard_callback = TensorBoard(log_dir=log_dir, histogram_freq=0, write_graph=False, write_images=False)
model.fit(x=train_images, y=train_labels, epochs=settings.epochs,
callbacks=[early_stopping_callback, model_checkpoint_callback, tensorboard_callback],
validation_data=(validation_images, validation_labels), batch_size=settings.batch_size,
verbose=verbose)
def prediction_for(predicted_value, cut_off, one_hot):
return predicted_value.max() < cut_off if one_hot else predicted_value < cut_off
def evaluate_model(model, settings, datasets):
column_headers = list(settings._fields)
column_values = list(settings)
cut_offs = [0.1 + 0.1 * i for i in range(8)]
for dataset in datasets:
column_headers.extend(["{}_{}".format(dataset.name, metric_name) for metric_name in model.metrics_names])
column_values.extend(model.evaluate(dataset.images, get_labels(dataset, settings.one_hot), settings.batch_size))
for cut_off in cut_offs:
column_headers.extend(["{}_{}_{}".format(dataset.name, cut_off, confusion_label)
for confusion_label in ["tn", "fp", "fn", "tp"]])
# Calculates confusion matrices for different cut-off at values
predicted_labels = [prediction_for(prediction, cut_off, settings.one_hot)
for prediction in model.predict(dataset.images)]
column_values.extend(confusion_matrix(dataset.labels, predicted_labels).ravel())
return column_headers, column_values
def train_and_evaluate(run_settings_list):
for run_settings in run_settings_list:
run_name = run_name_for(run_settings)
print("Training and evaluating: {}".format(run_name))
model = ModelFactory.model_for_settings(run_settings)
model.summary()
compiled_model = compile_model(model, run_settings)
dataset = Datasets.dataset_for(run_settings.dataset_name)
train_labels = get_labels(dataset[0], run_settings.one_hot)
test_labels = get_labels(dataset[2], run_settings.one_hot)
train_model(compiled_model, run_settings, dataset[0].images, train_labels, dataset[2].images, test_labels)
headers, values = evaluate_model(compiled_model, run_settings, dataset)
yield run_name, headers, values
def get_labels(dataset, one_hot):
return dataset.labels if not one_hot else to_categorical(dataset.labels)
def train_evaluate_and_log(csv_filename, run_settings_list):
write_header_row = True
run_names = []
column_headers = []
model_evaluations = []
with open(csv_filename, "w") as csv_file:
csv_writer = csv.writer(csv_file, delimiter=";")
for run_name, header, evaluations in train_and_evaluate(run_settings_list):
if write_header_row:
csv_writer.writerow(header)
write_header_row = False
values = [run_name]
values.extend(evaluations)
csv_writer.writerow(values)
csv_file.flush()
run_names.append(run_name)
column_headers.append(header)
model_evaluations.append(evaluations)
return run_names, column_headers, np.array(model_evaluations)
def load_run_settings(filename):
return [RunSettings(model_name="vgg16_gap", last_vgg_layer=last_layer, pre_trained_weights=weights, include_top=False, all_trainable=all_trainable,
dataset_name=dataset, batch_size=batch_size, epochs=epochs, optimizer="rmsprop",
lr=None, momentum=None,
decay=None, nesterov=None, one_hot=True)
for last_layer in ["block2_conv2", "block3_conv3", "block4_conv3", "block5_conv3"]
for dataset in list(Datasets.available_datasets())[:-1]
for weights in ["imagenet"]
for all_trainable in [False]
for epochs in [200]
for batch_size in [64]
]
# [RunSettings(model_name=model, last_vgg_layer="", pre_trained_weights=weights, include_top=False, all_trainable=all_trainable,
# dataset_name=dataset, batch_size=batch_size, epochs=epochs, optimizer="rmsprop",
# lr=None, momentum=None,
# decay=None, nesterov=None, one_hot=False)
# for model in ["vgg16", "vgg19", "xception", "resnet50"] #ModelFactory.available_base_models()
# for dataset in list(Datasets.available_datasets())[:-1]
# for weights in ["imagenet", None]
# for all_trainable in [False, True]
# for epochs in [200]
# for batch_size in [64]
# ]
def main(argv):
if len(argv) <= 1:
print("Usage: TrainModel.py <run settings filename>")
exit(0)
run_settings_list = load_run_settings(argv[1])
print(run_settings_list)
csv_filename = ProjectPaths.logfile_in_log_dir("all_runs_{}.csv")
run_names, column_headers, np_model_evaluations = train_evaluate_and_log(csv_filename, run_settings_list)
evaluations = pd.DataFrame(np_model_evaluations, index=run_names, columns=column_headers)
print(evaluations.head())
if __name__ == "__main__":
main(sys.argv)