-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgrid_run_best.py
173 lines (140 loc) · 7.07 KB
/
grid_run_best.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
import sys
import os
import numpy as np
import random
os.environ['PYTHONHASHSEED'] = str(76)
import tensorflow as tf
from tensorflow import keras
import time
import pandas as pd
from config import *
import glob
from astropy.io import fits
import data_utils
import plot_utils
import model_utils
import matplotlib.pyplot as plt
def keras_model_CC(filters_l1, filters_l3, filters_l5, neurons_l9, kernels_l1, kernels_l3, kernels_l5, dropout_l41, dropout_l82, learnrate):
"""
Parameters
Returns
- model: model to make later model.evaluate or model.predict
"""
random.seed(45)
np.random.seed(1)
tf.random.set_seed(346)
# -- define the network
layer1 = keras.layers.Conv2D(filters_l1, kernel_size=(kernels_l1, kernels_l1), padding="valid", activation="relu")
layer2 = keras.layers.MaxPooling2D((2, 2), strides=2)
layer3 = keras.layers.Conv2D(filters_l3, kernel_size=(kernels_l3, kernels_l3), padding="valid", activation="relu")
layer4 = keras.layers.MaxPooling2D((2, 2), strides=2)
layer41 = keras.layers.Dropout(dropout_l41)
layer5 = keras.layers.Conv2D(filters_l5, kernel_size=(kernels_l5, kernels_l5), padding="valid", activation="relu")
layer6 = keras.layers.MaxPooling2D((2, 2), strides=2)
layer82 = keras.layers.Dropout(dropout_l82)
layer7 = keras.layers.Flatten()
layer9 = keras.layers.Dense(neurons_l9, activation="relu")
layer10 = keras.layers.Dense(2, activation="softmax")
layers = [layer1,layer2,layer3,layer4,layer41,layer5,layer6,layer82,
layer7,layer9,layer10]
# -- instantiate the convolutional neural network
model = keras.Sequential(layers)
opt = keras.optimizers.SGD(learnrate)
model.compile(optimizer=opt, loss="sparse_categorical_crossentropy", metrics=["accuracy"])
#earlyst = tf.keras.callbacks.EarlyStopping(monitor="val_loss", mode="min", verbose=0, patience=100)
return model
# -- read data
data_location = "../data/data_split_3s/"
train, test, train_targ, test_targ, train_ID, test_ID = data_utils.read_data_folders(data_location, 10000, 2000, ddh = 1)
print(train.shape, test.shape)
print('unique test: {}'.format(np.unique(test_targ, return_counts=True)))
print('unique train: {}'.format(np.unique(train_targ, return_counts=True)))
# -- features need to have an extra axis on the end (for mini-batching)
feat_tr2 = train.reshape(len(train), train.shape[1], train.shape[2], 1)
feat_te2 = test.reshape(len(test), test.shape[1], test.shape[2], 1)
# -- initialize grid search with scikit-learn
#model = KerasClassifier(build_fn = keras_model_CC, epochs = 650, verbose = 0)
# -- set grid of hyperparameters
model_name = sys.argv[2]
params = sys.argv[1]
div = params.split("-")
filters_l1 = int(div[0])
filters_l3 = int(div[1])
filters_l5 = int(div[2])
neurons_l9 = int(div[3])
kernels_l1 = int(div[4])
kernels_l3 = int(div[5])
kernels_l5 = int(div[6])
dropout_l41 = float(div[7])
dropout_l82 = float(div[8])
learnrate = float(div[9])
batchsize = int(div[10])
model = keras_model_CC(filters_l1 = filters_l1, filters_l3 = filters_l3, filters_l5 = filters_l5, neurons_l9 = neurons_l9,
kernels_l1 = kernels_l1, kernels_l3 = kernels_l3, kernels_l5 = kernels_l5, dropout_l41 = dropout_l41, dropout_l82 = dropout_l82, learnrate = learnrate)
print("filers_l1: {}, filters_l3: {}, filters_l5 : {}, neurons_l9: {}, kernels_l1: {}, kernels_l3: {}, kernels_l5: {}, dropout_l41: {}, dropout_l82: {}, learnrate: {}, batchsize: {}".format(int(div[0]), int(div[1]), int(div[2]), int(div[3]), kernels_l1, kernels_l3, kernels_l5, float(div[7]), float(div[8]), float(div[9]), batchsize))
model_name = model_name+str(kernels_l1)+str(kernels_l3)+str(kernels_l5)+str(batchsize)
filepath = "model_checkpoint_%s.h5"%model_name
print(filepath)
checkpoint = tf.keras.callbacks.ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1,
save_best_only=True, mode='max')
earlyst = tf.keras.callbacks.EarlyStopping(monitor='val_loss', mode='min', verbose=0, patience=100)
print(train.shape, test.shape)
print('unique test: {}'.format(np.unique(test_targ, return_counts=True)))
print('unique train: {}'.format(np.unique(train_targ, return_counts=True)))
# -- feautres need to have an extra axis on the end (for mini-batching)
feat_tr2 = train.reshape(len(train), train.shape[1], train.shape[2], 1)
feat_te2 = test.reshape(len(test), test.shape[1], test.shape[2], 1)
times = 0
start = 0
end = 0
start = time.time()
# -- fit the model
checkpoints = 0
if checkpoints == 0:
history = model.fit(feat_tr2, train_targ, validation_split=0.20, epochs=650, batch_size=batchsize,
callbacks=[earlyst, checkpoint])
# -- specify if there is already a created model, and continue the training
else:
model = keras.models.load_model(filepath)
checkpoint = tf.keras.callbacks.ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1,
save_best_only=True, mode='max')
history = model.fit(feat_tr2, train_targ, validation_split=0.20, epochs=250, batch_size=200,
callbacks=[earlyst, checkpoint])
# -- print the accuracy
loss_tr, acc_tr = model.evaluate(feat_tr2, train_targ, verbose=0)
loss_te, acc_te = model.evaluate(feat_te2, test_targ, verbose=0)
print("Training accuracy : {0:.4f}".format(acc_tr))
print("Testing accuracy : {0:.4f}".format(acc_te))
end = time.time()
times = (end - start)
print(times)
pd.DataFrame(history.history["loss"]).to_csv("../outputs/trainc_loss%s"%model_name)
pd.DataFrame(history.history["val_loss"]).to_csv("../outputs/testc_loss%s"%model_name)
pd.DataFrame(history.history["accuracy"]).to_csv("../outputs/trainc_acc%s"%model_name)
pd.DataFrame(history.history["val_accuracy"]).to_csv("../outputs/testc_acc%s"%model_name)
fig, ax = plt.subplots(2,1,figsize=(8, 8))
ax[0].plot(history.history["loss"], color = "#00441B")
ax[0].plot(history.history["val_loss"], color = "#40004B")
ax[0].legend(["train", "validation"], loc="upper right")
ax[0].set_xlabel("Epoch", fontsize=15)
ax[0].set_xticks([])
ax[0].set_ylabel("Loss", fontsize=15)
ax[1].plot(history.history["accuracy"], color = "#00441B")
ax[1].plot(history.history["val_accuracy"], color = "#40004B")
ax[1].legend(["train", "validation"], loc="lower right")
ax[1].set_xlabel("Epoch", fontsize=15)
ax[1].set_ylabel("Accuracy", fontsize=15)
if checkpoints == 0:
plt.savefig("../outputs/lossacc_model%s.pdf"%model_name,bbox_inches="tight")
else:
plt.savefig("../outputs/lossacc_model%s_checkpoint.pdf"%model_name, bbox_inches="tight")
y_pred_test = model_utils.predict_data(feat_te2, model, 0, test_ID,test_targ)
if checkpoints == 0:
plot_utils.create_confusion_matrix("test_%s"%model_name, y_pred_test, test_targ)
else:
plot_utils.create_confusion_matrix("test_%s_checkpoint"%model_name, y_pred_test, test_targ)
y_pred_train = model_utils.predict_data(feat_tr2, model, 0, train_ID, train_targ)
if checkpoints == 0:
plot_utils.create_confusion_matrix("train_%s"%model_name, y_pred_train, train_targ)
else:
plot_utils.create_confusion_matrix("train_%s_checkpoint"%model_name, y_pred_train, train_targ)