-
Notifications
You must be signed in to change notification settings - Fork 1
/
ICAlib.py
207 lines (180 loc) · 7.22 KB
/
ICAlib.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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
import tensorflow as tf
import numpy as np
from sklearn.decomposition import PCA
import tensorflow.contrib.distributions as tfd
from munkres import Munkres
NP_DTYPE = np.float32
DTYPE= tf.float32
NPC = np.log2(np.exp(1))
activations = dict()
activations['tanh']=tf.tanh
activations['tan']=tf.tan
activations['sin']=tf.sin
activations['cos']=tf.cos
activations['exp']=tf.exp
activations['logabs']=lambda x:tf.log( tf.abs(x)+1e-10)
activations['abs']=tf.abs
activations['myhardtanh']=lambda x: tf.nn.leaky_relu( 1.0- tf.nn.leaky_relu(1.0-x) )
activations['softplus']=tf.nn.softplus
activations['relu']=tf.nn.relu
activations['lrelu']=tf.nn.leaky_relu
activations['relu6']=tf.nn.relu6
activations['elu']=tf.nn.elu
activations['crelu']=tf.nn.crelu
activations['selu']=tf.nn.selu
activations['softsign']=tf.nn.softsign
activations['sigmoid']=tf.nn.sigmoid
activations['hardsigmoid']=tf.keras.activations.hard_sigmoid
activations['identity']=tf.identity
def get_nb_params_shape(shape):
nb_params = 1
for dim in shape:
nb_params = nb_params*int(dim)
return nb_params
def count_number_trainable_params():
'''
Counts the number of trainable variables.
'''
tot_nb_params = 0
for trainable_variable in tf.trainable_variables():
shape = trainable_variable.get_shape() # e.g [D,F] or [W,H,C]
current_nb_params = get_nb_params_shape(shape)
tot_nb_params = tot_nb_params + current_nb_params
return tot_nb_params
def weight_variable(shape, stddev=0.01, name='weight'):
initial = tf.truncated_normal(shape, stddev=stddev)
return tf.Variable(initial, name=name)
def bias_variable(shape, value=0.0, name='bias'):
initial = tf.constant(value, shape=shape)
return tf.Variable(initial, name=name)
def get_data(seed=101, mix_dim=6, task_type='linear', samples=4000,noise_std=0,whiten=True):
np.random.seed(seed)
t = np.linspace(0, samples * 1e-4, samples)
two_pi = 2 * np.pi
s0 = np.sign(np.cos(two_pi * 155 * t))
s1 = np.sin(two_pi * 800 * t)
s2 = np.sin(two_pi * 300 * t + 6 * np.cos(two_pi * 60 * t))
s3 = np.sin(two_pi * 90 * t)
s4 = np.random.uniform(-1, 1, (samples,))
s5 = np.random.laplace(0, 1, (samples,))
x = np.stack([s0, s1, s2, s3, s4, s5])
mix_mat = np.random.uniform(-.5, .5, (mix_dim, 6))
y = np.dot(mix_mat, x)
if task_type in ['sin']:
y = np.sin(y)
if task_type in ['cos']:
y = np.cos(y)
if task_type in ['axbx2']:
mix_mat2 = np.random.uniform(-.5, .5, (mix_dim, mix_dim))
y1 = np.dot(mix_mat2, x*x*x)
y = y+y1
if task_type in ['mlp', 'pnl']:
y = np.tanh(y)
if task_type in ['mlp', 'mlp3']:
mix_mat2 = np.random.uniform(-.5, .5, (mix_dim, mix_dim))
y = np.tanh(np.dot(mix_mat2, y))
if task_type in ['mlp3', 'mlp4']:
mix_mat3 = np.random.uniform(-.5, .5, (mix_dim, mix_dim))
y = np.tanh(np.dot(mix_mat3, y))
if noise_std>0:
y = y + np.random.multivariate_normal([0]*6,noise_std*np.eye(6), (samples,)).T
pca = PCA(whiten=True)
if whiten:
y=pca.fit_transform(y.T)
y=y.T
return x.T, y.T ,pca
def FC(inputs,sizes,activations=None,name='fc',ki=None):
if isinstance(inputs,(list,tuple)):
inputs = tf.concat(inputs,-1)
if not isinstance(sizes,(list,tuple)):
sizes=[sizes]
X = inputs
if not isinstance(activations,(list,tuple)):
activations = [activations] * len(sizes)
for i in range(len(sizes)):
if len(activations) < i-1:
act = None
else:
act = activations[i]
X = tf.layers.dense( X, sizes[i] , act , name=name+'_%d'%(i+1) ,kernel_initializer=ki)
return X
def create_gmm_1( d,K,name='gmm', reuse=False, scale_act=tf.nn.softplus,zero_mean=False,ki=None):
with tf.variable_scope( name , reuse):
#tf.random_uniform_initializer(0.,3.)
probs = tf.nn.softmax( tf.get_variable('probs', shape=[d,K] , dtype=DTYPE,initializer=None),axis=-1)
#tf.random_uniform_initializer(-.5,.5)
locs = tf.get_variable('locs', shape=[d,K] , dtype=DTYPE,initializer=None)
if zero_mean:
locs=tf.zeros_like(locs)
scales = tf.get_variable('scales', shape=[d,K] , dtype=DTYPE,initializer=None)
pis = tfd.Categorical( probs = probs )
ps = tfd.Normal( loc = locs , scale =scale_act(scales))
p = tf.contrib.distributions.MixtureSameFamily( pis, ps)
p = tf.contrib.distributions.Independent(p,1)
return p
def get_random_batch(x, n):
indices = np.random.randint(x.shape[0], size=(n,))
return x[indices]
def mycorrelation(x, y, method='Pearson'):
"""Evaluate correlation
Args:
x: data to be sorted
y: target data
Returns:
corr_sort: correlation matrix between x and y (after sorting)
sort_idx: sorting index
x_sort: x after sorting
method: correlation method ('Pearson' or 'Spearman')
"""
# print("Calculating correlation...")
x = x.copy()
y = y.copy()
dim = x.shape[0]
# Calculate correlation -----------------------------------
if method=='Pearson':
corr = np.corrcoef(y, x)
corr = corr[0:dim,dim:]
elif method=='Spearman':
corr, pvalue = sp.stats.spearmanr(y.T, x.T)
corr = corr[0:dim, dim:]
# Sort ----------------------------------------------------
munk = Munkres()
indexes = munk.compute(-np.absolute(corr))
sort_idx = np.zeros(dim)
x_sort = np.zeros(x.shape)
for i in range(dim):
sort_idx[i] = indexes[i][1]
x_sort[i,:] = x[indexes[i][1],:]
# Re-calculate correlation --------------------------------
if method=='Pearson':
corr_sort = np.corrcoef(y, x_sort)
corr_sort = corr_sort[0:dim,dim:]
elif method=='Spearman':
corr_sort, pvalue = sp.stats.spearmanr(y.T, x_sort.T)
corr_sort = corr_sort[0:dim, dim:]
return corr_sort, sort_idx, x_sort
def sample_integers(n, shape):
sample = tf.random_uniform(shape, minval=0, maxval=tf.cast(n, 'float32'))
sample = tf.cast(sample, 'int32')
return sample
def resample_rows_per_column(x):
"""Permute all rows for each column independently."""
n_batch = tf.shape(x)[0]
n_dim = tf.shape(x)[1]
row_indices = sample_integers(n_batch, (n_batch * n_dim,))
col_indices = tf.tile(tf.range(n_dim), [n_batch])
indices = tf.transpose(tf.stack([row_indices, col_indices]))
x_perm = tf.gather_nd(x, indices)
x_perm = tf.reshape(x_perm, (n_batch, n_dim))
return x_perm
def get_gan_cost(logits):
return tf.reduce_mean(tf.nn.softplus(logits))
def cross_entropy_with_logits_loss(logits, targets):
log_q = tf.nn.log_softmax(logits)
return -tf.reduce_sum(targets * log_q, 1)
def tf_cov(x):
mean_x = tf.reduce_mean(x, axis=0, keep_dims=True)
mx = tf.matmul(tf.transpose(mean_x), mean_x)
vx = tf.matmul(tf.transpose(x), x)/tf.cast(tf.shape(x)[0], tf.float32)
cov_xx = vx - mx
return cov_xx