deepDR/getFeatures.py at master · ChengF-Lab/deepDR · 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
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
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import minmax_scale, scale
from keras.models import Model, load_model
from keras.callbacks import EarlyStopping
from MDA import build_MDA, build_AE, build_MDA2

import matplotlib as mpl
mpl.use('Agg')
import matplotlib.pyplot as plt
import os.path as Path
import scipy.io as sio
import numpy as np
import pickle
import sys


def read_params(fname):
    params = {}
    fR = open(fname, 'r')
    for line in fR:
        print (line.strip())
        key, val = line.strip().split('=')
        key = str(key.strip())
        val = str(val.strip())
        if key == 'select_arch' or key == 'select_nets':
            params[key] = map(int, val.strip('[]').split(','))
        else:
            params[key] = str(val)
    print ("###############################################################")
    print
    print
    fR.close()

    return params


def build_model(X, input_dims, arch, nf=0.5, std=1.0, mtype='mda', epochs=80, batch_size=64):
    if mtype == 'mda':
        model = build_MDA(input_dims, arch)
        # model = build_MDA2(input_dims, arch)
    elif mtype == 'ae':
        model = build_AE(input_dims[0], arch)
    else:
        print ("### Wrong model.")
    # corrupting the input
    noise_factor = nf
    if isinstance(X, list):
        Xs = train_test_split(*X, test_size=0.2)
        X_train = []
        X_test = []
        for jj in range(0, len(Xs), 2):
            X_train.append(Xs[jj])
            X_test.append(Xs[jj+1])
        X_train_noisy = list(X_train)
        X_test_noisy = list(X_test)
        for ii in range(0, len(X_train)):
            X_train_noisy[ii] = X_train_noisy[ii] + noise_factor*np.random.normal(loc=0.0, scale=std, size=X_train[ii].shape)
            X_test_noisy[ii] = X_test_noisy[ii] + noise_factor*np.random.normal(loc=0.0, scale=std, size=X_test[ii].shape)
            X_train_noisy[ii] = np.clip(X_train_noisy[ii], 0, 1)
            X_test_noisy[ii] = np.clip(X_test_noisy[ii], 0, 1)
    else:
        X_train, X_test = train_test_split(X, test_size=0.2)
        X_train_noisy = X_train.copy()
        X_test_noisy = X_test.copy()
        X_train_noisy = X_train_noisy + noise_factor*np.random.normal(loc=0.0, scale=std, size=X_train.shape)
        X_test_noisy = X_test_noisy + noise_factor*np.random.normal(loc=0.0, scale=std, size=X_test.shape)
        X_train_noisy = np.clip(X_train_noisy, 0, 1)
        X_test_noisy = np.clip(X_test_noisy, 0, 1)
    # Fitting the model
    history = model.fit(X_train_noisy, X_train, epochs=epochs, batch_size=batch_size, shuffle=True,
                        validation_data=(X_test_noisy, X_test),
                        callbacks=[EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=5)])
    mid_model = Model(inputs=model.input, outputs=model.get_layer('middle_layer').output)

    return mid_model, history


### Main code starts here
params = read_params(sys.argv[1])


org = params['org']   # {drug}
valid_type = params['valid_type']  # {cv}
model_type = params['model_type']  # {mda or ae}
ofile_keywords = params['ofile_keywords']  # {example: 'final_res'}

models_path = params['models_path']  # directory with models
results_path = params['results_path']  # directotry with results
select_arch = params['select_arch']  # a number 1-10 (see below)
select_nets = params['select_nets']  # a number 1-10 (see below)
epochs = int(params['epochs'])
batch_size = int(params['batch_size'])
nf = float(params['noise_factor'])  # nf > 0 for denoising AE/MDA
n_trials = int(params['n_trials'])  # number of cv trials
K = params['K']  # number of propagations
alpha = params['alpha']  # propagation parameter


# all possible combinations for architectures
arch = {}
arch['mda'] = {}
arch['mda']['drug'] = {}
arch['mda']['drug'] = {1: [9*100],
                        2: [9*1000, 9*100, 9*1000],
                        3: [9*1000, 9*500, 9*100, 9*500, 9*1000],
                        4: [9*1000, 9*500, 9*200, 9*100, 9*200, 9*500, 9*1000],
                        5: [9*1000, 9*800, 9*500, 9*200, 9*100, 9*200, 9*500, 9*800, 9*1000],
                        }

arch['ae'] = {}
arch['ae']['drug'] = {}
arch['ae']['drug'] = {1: [1000],
                       2: [2000, 1000, 2000],
                       3: [2000, 1500, 1000, 1500, 2000],
                       }

# load PPMI matrices
Nets = []
input_dims = []
for i in select_nets:
    print ("### [%d] Loading network..." % (i))
    N = sio.loadmat('./PPMI/' + org + '_net_' + str(i) +  '.mat', squeeze_me=True)
    Net = N['Net'].todense()
    print ("Net %d, NNofile_keywords=%d \n" % (i, np.count_nonzero(Net)))
    Nets.append(minmax_scale(Net))
    input_dims.append(Net.shape[1])


# Training MDA/AE
model_names = []
for a in select_arch:
    print ("### [%s] Running for architecture: %s" % (model_type, str(arch[model_type][org][a])))
    model_name = org + '_' + model_type.upper() + '_arch_' + str(a) + '_' + ofile_keywords + '.h5'
    if not Path.isfile(models_path + model_name):
        mid_model, history = build_model(Nets, input_dims, arch[model_type][org][a], nf, 1.0, model_type, epochs, batch_size)
        # save model
        mid_model.save(models_path + model_name)
        with open(models_path + model_name.split('.')[0] + '_history.pckl', 'wb') as file_pi:
            pickle.dump(history.history, file_pi)

        # Export figure: loss vs epochs (history)
        plt.figure()
        plt.plot(history.history['loss'], '.-')
        plt.plot(history.history['val_loss'], '.-')
        plt.title('model loss')
        plt.ylabel('loss')
        plt.xlabel('epoch')
        plt.legend(['train', 'validation'], loc='upper left')
        plt.savefig(models_path + model_name + '_loss.png', bbox_inches='tight')
    model_names.append(model_name)


# Saving features
for model_name in model_names:
    print ("### Running for: %s" % (model_name))
    if Path.isfile(models_path + model_name):
        mid_model = load_model(models_path + model_name)
    else:
        print ("### Model % s does not exist. Check the 'models_path' directory.\n" % (model_name))
        break
    mid_model = load_model(models_path + model_name)
    features = mid_model.predict(Nets)
    features = minmax_scale(features)
    # sio.savemat(models_path + model_name.split('.')[0] + '_features.mat', {'features': features})
    np.savetxt(org + model_type+ 'Features.txt', features, delimiter='\t', fmt='%s',newline='\n')
    print ("Done!")