master-imagedataanalysis/TrainMIp(1).py at main · aysenurdorr-gif/master-imagedataanalysis · GitHub

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
from keras.models import Sequential
import scipy.io
import numpy
import sys
from keras.layers.core import Dense, Activation
from keras.optimizers import SGD, RMSprop, Adagrad, Adadelta, Adam, Nadam
import h5py
# from  normalize_data import normalize,denormalize
from sklearn.cross_validation import train_test_split
from keras.utils.generic_utils import Progbar
from keras.callbacks import EarlyStopping
import os
from keras.callbacks import TensorBoard
import ntpath
import logging
from keras.callbacks import ModelCheckpoint
import glob, os

cnfg = sys.argv[1];
datafile = sys.argv[2];
mdlfile = sys.argv[3];
pred_path = sys.argv[4];
test_path = sys.argv[5];
# train_flag = int(sys.argv[4]);
newDict = {}

with open(cnfg) as f:
    for line in f:
        line = line.rstrip('\n')
        print(line)
        splitLine = line.split(':')
        if ((splitLine[0] == 'indim') | (splitLine[0] == 'outdim') | (splitLine[0] == 'nbepochs') | (
                splitLine[0] == 'batchsize') | (splitLine[0] == 'normalizex') | (splitLine[0] == 'normalizey')):
            newDict[(splitLine[0])] = int(splitLine[1].rstrip())
        elif (splitLine[0] == 'loss'):
            newDict[(splitLine[0])] = (splitLine[1].strip())
        else:
            line = splitLine[1];
            line = line.split(' ')
            newDict[(splitLine[0])] = line

#    if(splitLine[0]!='nbepochs'):
# mdlfile = mdlfile+splitLine[0]+'_'+ str(newDict[(splitLine[0])])

# name=os.path.splitext(datafile)[0]
# C= name.split('/')
# print C[-1],C[-2]
# mdlfile = C[-1]+'_'+C[-2]+'_'+xname+'_'+yname;
directory = './models/' + mdlfile;
logging.basicConfig(level=logging.DEBUG)

print(directory)

os.system('cp ' + cnfg + ' ' + directory + '/nnet_config.txt')

print(newDict)
# nins = newDict['indim']
# nouts = newDict['outdim']
activations = newDict['activation']

hiddenlayers = newDict['hiddenlayers']
hiddenlayers = [x for x in hiddenlayers if (x != '')]

data = scipy.io.loadmat(Data);
Xtrain = data['train_data']
Ytrain = data['train_lab']

Xval = data['val_data']
Yval = data['val_lab']

Xtest = data['test_data']
Ytest = data['test_lab']

# Ytrain = numpy.log(Ytrain)
print(Xtrain.shape)
if (newDict['normalizex'] == 1):
    print("normalizing input data")
    mux = numpy.mean(Xtrain, axis=0);
    sigmax = numpy.std(Xtrain, axis=0);
    print(mux.shape, sigmax.shape)
    sigmax[sigmax < 1e-5] = 1;
    numpy.savetxt(directory + '/mux.txt', mux);
    numpy.savetxt(directory + '/sigmax.txt', sigmax);
    Xtrain = (Xtrain - mux) / sigmax;
    Xval = (Xval - mux) / sigmax;
    Xtest = (Xtest - mux) / sigmax;

print('number of infinite values:' + str(sum(sum(~numpy.isfinite(Xtrain)))))
xtrain_size = Xtrain.shape
ytrain_size = Ytrain.shape
print('input number of greater than +-1:' + str(sum(sum(abs(Xtrain) > 1)) * 100.00 / (xtrain_size[0] * xtrain_size[1])))
print('input data dimention is ' + str(Xtrain.shape))
model = Sequential()
print('..building model')
print('creating the layer: Input {} -> Output {} with activation {}'.format(Xtrain.shape[1], int(hiddenlayers[1]),
                                                                            activations[0]))
model.add(Dense(output_dim=int(hiddenlayers[1]), input_dim=Xtrain.shape[1], activation=activations[0]))
for k in xrange(2, len(hiddenlayers) - 1):
    print('creating the layer: Input {} -> Output {} with activation {}'.format(int(hiddenlayers[k - 1]),
                                                                                int(hiddenlayers[k]), activations[k]))
    model.add(Dense(output_dim=int(hiddenlayers[k]), activation=activations[k]))

print('creating the layer: Input {} -> Output {} with activation {}'.format(int(hiddenlayers[len(hiddenlayers) - 2]),
                                                                            Ytrain.shape[1], activations[-1]))
model.add(Dense(output_dim=int(Ytrain.shape[1]), activation=activations[-1]))

print('..compiling model')
model.compile(loss=newDict['loss'], optimizer=Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08),
              metrics=['accuracy'])

print('..fitting model')
batch_size = int(newDict['batchsize'])
early_stopping = EarlyStopping(monitor='val_loss', patience=2)
checkpointer = ModelCheckpoint(filepath='saved_weight.hdf5', verbose=1, save_best_only=True)
history2 = model.fit(Xtrain, Ytrain, nb_epoch=int(newDict['nbepochs']), batch_size=int(newDict['batchsize']),
                     validation_data=(Xval, Yval), callbacks=[early_stopping, checkpointer])
model.load_weights('saved_weight.hdf5')
list1 = glob.glob(test_path + "*.mat");

for k in list1:
    print('processing ' + k)
    st = k.split('/');
    name = st[-1]
    data = scipy.io.loadmat(k)
    Xtest = data['all_data'].transpose();
    Xtest = (Xtest - mux) / sigmax;
    pred = model.predict(Xtest, verbose=1);
    scipy.io.savemat(pred_path + name, {'pred': pred})