-
Notifications
You must be signed in to change notification settings - Fork 6
/
main_clustering.py
381 lines (323 loc) · 18.4 KB
/
main_clustering.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
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
from __future__ import division
import os,sys
import time
import dateutil.tz
import datetime
import argparse
import importlib
import tensorflow as tf
import numpy as np
import random
import copy
import math
import util
import metric
from sklearn.cluster import KMeans
from sklearn.metrics.cluster import normalized_mutual_info_score, adjusted_rand_score,homogeneity_score
import pandas as pd
tf.set_random_seed(0)
tf.reset_default_graph()
'''
Instructions: scDEC model
x,y - data drawn from base density (e.g., Gaussian) and observation data
x_onehot - data drawn from caltegrory distribution
y_ - Generated data where y_=G(x,x_onehot)
x_latent_,x_onehot_ - Embedding and inferred clustering label where x_latent_, x_onehot_=H(y)
y__ - reconstructed distribution, y__ = G(H(y))
x__ - reconstructed distribution, x__ = H(G(y))
G(.) - generator network for mapping latent space to data space
H(.) - generator network for mapping data space to latent space (embedding) and clustering, simultaneously
Dx(.) - discriminator network in x space (latent space)
Dy(.) - discriminator network in y space (observation space)
'''
class scDEC(object):
def __init__(self, g_net, h_net, dx_net, dy_net, x_sampler, y_sampler, nb_classes, data, pool, batch_size, alpha, beta, is_train):
self.data = data
self.g_net = g_net
self.h_net = h_net
self.dx_net = dx_net
self.dy_net = dy_net
self.x_sampler = x_sampler
self.y_sampler = y_sampler
self.nb_classes = nb_classes
self.batch_size = batch_size
self.alpha = alpha
self.beta = beta
self.pool = pool
self.x_dim = self.dx_net.input_dim
self.y_dim = self.dy_net.input_dim
self.x = tf.placeholder(tf.float32, [None, self.x_dim], name='x')
self.x_onehot = tf.placeholder(tf.float32, [None, self.nb_classes], name='x_onehot')
self.x_combine = tf.concat([self.x,self.x_onehot],axis=1,name='x_combine')
self.y = tf.placeholder(tf.float32, [None, self.y_dim], name='y')
self.y_ = self.g_net(self.x_combine,reuse=False)
self.x_latent_, self.x_onehot_ = self.h_net(self.y,reuse=False)#continuous + softmax + before_softmax
self.x_ = self.x_latent_[:,:self.x_dim]
self.x_logits_ = self.x_latent_[:,self.x_dim:]
self.x_latent__, self.x_onehot__ = self.h_net(self.y_)
self.x__ = self.x_latent__[:,:self.x_dim]
self.x_logits__ = self.x_latent__[:,self.x_dim:]
self.x_combine_ = tf.concat([self.x_, self.x_onehot_],axis=1)
self.y__ = self.g_net(self.x_combine_)
self.dy_ = self.dy_net(self.y_, reuse=False)
self.dx_ = self.dx_net(self.x_, reuse=False)
self.l2_loss_x = tf.reduce_mean((self.x - self.x__)**2)
self.l2_loss_y = tf.reduce_mean((self.y - self.y__)**2)
#self.CE_loss_x = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=self.x_onehot, logits=self.x_logits__))
self.CE_loss_x = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=self.x_logits__,labels=self.x_onehot))
self.g_loss_adv = -tf.reduce_mean(self.dy_)
self.h_loss_adv = -tf.reduce_mean(self.dx_)
self.g_loss = self.g_loss_adv + self.alpha*self.l2_loss_x + self.beta*self.l2_loss_y
self.h_loss = self.h_loss_adv + self.alpha*self.l2_loss_x + self.beta*self.l2_loss_y
self.g_h_loss = self.g_loss_adv + self.h_loss_adv + self.alpha*(self.l2_loss_x + self.l2_loss_y) + self.beta*self.CE_loss_x
self.dx = self.dx_net(self.x)
self.dy = self.dy_net(self.y)
self.dx_loss = -tf.reduce_mean(self.dx) + tf.reduce_mean(self.dx_)
self.dy_loss = -tf.reduce_mean(self.dy) + tf.reduce_mean(self.dy_)
#gradient penalty for x
epsilon_x = tf.random_uniform([], 0.0, 1.0)
x_hat = epsilon_x * self.x + (1 - epsilon_x) * self.x_
dx_hat = self.dx_net(x_hat)
grad_x = tf.gradients(dx_hat, x_hat)[0] #(bs,x_dim)
grad_norm_x = tf.sqrt(tf.reduce_sum(tf.square(grad_x), axis=1))#(bs,)
self.gpx_loss = tf.reduce_mean(tf.square(grad_norm_x - 1.0))
#gradient penalty for y
epsilon_y = tf.random_uniform([], 0.0, 1.0)
y_hat = epsilon_y * self.y + (1 - epsilon_y) * self.y_
dy_hat = self.dy_net(y_hat)
grad_y = tf.gradients(dy_hat, y_hat)[0] #(bs,x_dim)
grad_norm_y = tf.sqrt(tf.reduce_sum(tf.square(grad_y), axis=1))#(bs,)
self.gpy_loss = tf.reduce_mean(tf.square(grad_norm_y - 1.0))
self.d_loss = self.dx_loss + self.dy_loss + 10*(self.gpx_loss + self.gpy_loss)
self.lr = tf.placeholder(tf.float32, None, name='learning_rate')
self.g_h_optim = tf.train.AdamOptimizer(learning_rate=self.lr, beta1=0.5, beta2=0.9) \
.minimize(self.g_h_loss, var_list=self.g_net.vars+self.h_net.vars)
#self.d_optim = tf.train.GradientDescentOptimizer(learning_rate=self.lr) \
# .minimize(self.d_loss, var_list=self.dx_net.vars+self.dy_net.vars)
self.d_optim = tf.train.AdamOptimizer(learning_rate=self.lr, beta1=0.5, beta2=0.9) \
.minimize(self.d_loss, var_list=self.dx_net.vars+self.dy_net.vars)
now = datetime.datetime.now(dateutil.tz.tzlocal())
self.timestamp = now.strftime('%Y%m%d_%H%M%S')
self.g_loss_adv_summary = tf.summary.scalar('g_loss_adv',self.g_loss_adv)
self.h_loss_adv_summary = tf.summary.scalar('h_loss_adv',self.h_loss_adv)
self.l2_loss_x_summary = tf.summary.scalar('l2_loss_x',self.l2_loss_x)
self.l2_loss_y_summary = tf.summary.scalar('l2_loss_y',self.l2_loss_y)
self.dx_loss_summary = tf.summary.scalar('dx_loss',self.dx_loss)
self.dy_loss_summary = tf.summary.scalar('dy_loss',self.dy_loss)
self.gpx_loss_summary = tf.summary.scalar('gpx_loss',self.gpx_loss)
self.gpy_loss_summary = tf.summary.scalar('gpy_loss',self.gpy_loss)
self.g_merged_summary = tf.summary.merge([self.g_loss_adv_summary, self.h_loss_adv_summary,\
self.l2_loss_x_summary,self.l2_loss_y_summary,self.gpx_loss_summary,self.gpy_loss_summary])
self.d_merged_summary = tf.summary.merge([self.dx_loss_summary,self.dy_loss_summary])
#graph path for tensorboard visualization
self.graph_dir = 'graph/{}/{}_x_dim={}_y_dim={}_alpha={}_beta={}_ratio={}'.format(self.data,self.timestamp,self.x_dim, self.y_dim, self.alpha, self.beta, ratio)
if not os.path.exists(self.graph_dir) and is_train:
os.makedirs(self.graph_dir)
#save path for saving predicted data
self.save_dir = 'results/{}/{}_x_dim={}_y_dim={}_alpha={}_beta={}_ratio={}'.format(self.data,self.timestamp,self.x_dim, self.y_dim, self.alpha, self.beta, ratio)
if not os.path.exists(self.save_dir) and is_train:
os.makedirs(self.save_dir)
self.saver = tf.train.Saver(max_to_keep=5000)
#run_config = tf.ConfigProto(intra_op_parallelism_threads=1,inter_op_parallelism_threads=1)
run_config = tf.ConfigProto()
run_config.gpu_options.per_process_gpu_memory_fraction = 1.0
run_config.gpu_options.allow_growth = True
self.sess = tf.Session(config=run_config)
def train(self, nb_batches):
data_y = self.y_sampler.load_all()[0] if has_label else self.y_sampler.load_all()
self.sess.run(tf.global_variables_initializer())
self.summary_writer=tf.summary.FileWriter(self.graph_dir,graph=tf.get_default_graph())
batches_per_eval = 100
start_time = time.time()
weights = np.ones(self.nb_classes, dtype=np.float64) / float(self.nb_classes)
last_weights = np.ones(self.nb_classes, dtype=np.float64) / float(self.nb_classes)
diff_history=[]
for batch_idx in range(nb_batches):
lr = 2e-4
#update D
for _ in range(5):
bx, bx_onehot = self.x_sampler.train(self.batch_size,weights)
by = random.sample(data_y,self.batch_size)
d_summary,_ = self.sess.run([self.d_merged_summary, self.d_optim], feed_dict={self.x: bx, self.x_onehot: bx_onehot, self.y: by, self.lr:lr})
self.summary_writer.add_summary(d_summary,batch_idx)
bx, bx_onehot = self.x_sampler.train(self.batch_size,weights)
by = random.sample(data_y,self.batch_size)
#update G
g_summary, _ = self.sess.run([self.g_merged_summary ,self.g_h_optim], feed_dict={self.x: bx, self.x_onehot: bx_onehot, self.y: by, self.lr:lr})
self.summary_writer.add_summary(g_summary,batch_idx)
#quick test on a random batch data
if batch_idx % batches_per_eval == 0:
g_loss_adv, h_loss_adv, CE_loss, l2_loss_x, l2_loss_y, g_loss, \
h_loss, g_h_loss, gpx_loss, gpy_loss = self.sess.run(
[self.g_loss_adv, self.h_loss_adv, self.CE_loss_x, self.l2_loss_x, self.l2_loss_y, \
self.g_loss, self.h_loss, self.g_h_loss, self.gpx_loss, self.gpy_loss],
feed_dict={self.x: bx, self.x_onehot: bx_onehot, self.y: by}
)
dx_loss, dy_loss, d_loss = self.sess.run([self.dx_loss, self.dy_loss, self.d_loss], \
feed_dict={self.x: bx, self.x_onehot: bx_onehot, self.y: by})
print('Batch_idx [%d] Time [%.4f] g_loss_adv [%.4f] h_loss_adv [%.4f] CE_loss [%.4f] gpx_loss [%.4f] gpy_loss [%.4f] \
l2_loss_x [%.4f] l2_loss_y [%.4f] g_loss [%.4f] h_loss [%.4f] g_h_loss [%.4f] dx_loss [%.4f] dy_loss [%.4f] d_loss [%.4f]' %
(batch_idx, time.time() - start_time, g_loss_adv, h_loss_adv, CE_loss, gpx_loss, gpy_loss, l2_loss_x, l2_loss_y, \
g_loss, h_loss, g_h_loss, dx_loss, dy_loss, d_loss))
if (batch_idx+1) % batches_per_eval == 0:
self.evaluate(timestamp,batch_idx)
self.save(batch_idx)
tol = 0.02
estimated_weights = self.estimate_weights(use_kmeans=False)
weights = ratio*weights + (1-ratio)*estimated_weights
weights = weights/np.sum(weights)
diff_weights = np.mean(np.abs(last_weights-weights))
diff_history.append(diff_weights)
if np.min(weights)<tol:
weights = self.adjust_tiny_weights(weights,tol)
last_weights = copy.copy(weights)
if len(diff_history)>100 and np.mean(diff_history[-10:]) < 5e-3 and batch_idx>30000:
print('Reach a stable cluster')
self.evaluate(timestamp,batch_idx)
sys.exit()
def adjust_tiny_weights(self,weights,tol):
idx_less = np.where(weights<tol)[0]
idx_greater = np.where(weights>=tol)[0]
weights[idx_less] = np.array([np.random.uniform(2*tol,1./self.nb_classes) for item in idx_less])
weights[idx_greater] = weights[idx_greater]*(1-np.sum(weights[idx_less]))/np.sum(weights[idx_greater])
return weights
def estimate_weights(self,use_kmeans=False):
data_y = self.y_sampler.load_all()[0] if has_label else self.y_sampler.load_all()
data_x_, data_x_onehot_ = self.predict_x(data_y)
if use_kmeans:
km = KMeans(n_clusters=nb_classes, random_state=0).fit(np.concatenate([data_x_,data_x_onehot_],axis=1))
label_infer = km.labels_
else:
label_infer = np.argmax(data_x_onehot_, axis=1)
weights = np.empty(self.nb_classes, dtype=np.float32)
for i in range(self.nb_classes):
weights[i] = list(label_infer).count(i)
return weights/float(np.sum(weights))
def evaluate(self,timestamp,batch_idx):
if has_label:
data_y, label_y = self.y_sampler.load_all()
else:
data_y = self.y_sampler.load_all()
data_x_, data_x_onehot_ = self.predict_x(data_y)
label_infer = np.argmax(data_x_onehot_, axis=1)
if has_label:
purity = metric.compute_purity(label_infer, label_y)
nmi = normalized_mutual_info_score(label_y, label_infer)
ari = adjusted_rand_score(label_y, label_infer)
homo = homogeneity_score(label_y,label_infer)
print('scDEC: NMI = {}, ARI = {}, Homogeneity = {}'.format(nmi,ari,homo))
if is_train:
f = open('%s/log.txt'%self.save_dir,'a+')
f.write('NMI = {}\tARI = {}\tHomogeneity = {}\t batch_idx = {}\n'.format(nmi,ari,homo,batch_idx))
f.close()
np.savez('{}/data_at_{}.npz'.format(self.save_dir, batch_idx+1),data_x_,data_x_onehot_,label_y)
else:
np.savez('results/{}/data_pre.npz'.format(self.data),data_x_,data_x_onehot_,label_y)
else:
if is_train:
np.savez('{}/data_at_{}.npz'.format(self.save_dir, batch_idx+1),data_x_,data_x_onehot_)
else:
np.savez('results/{}/data_pre.npz'.format(self.data),data_x_,data_x_onehot_)
#predict with y_=G(x)
def predict_y(self, x, x_onehot, bs=256):
assert x.shape[-1] == self.x_dim
N = x.shape[0]
y_pred = np.zeros(shape=(N, self.y_dim))
for b in range(int(np.ceil(N*1.0 / bs))):
if (b+1)*bs > N:
ind = np.arange(b*bs, N)
else:
ind = np.arange(b*bs, (b+1)*bs)
batch_x = x[ind, :]
batch_x_onehot = x_onehot[ind, :]
batch_y_ = self.sess.run(self.y_, feed_dict={self.x:batch_x, self.x_onehot:batch_x_onehot})
y_pred[ind, :] = batch_y_
return y_pred
#predict with x_=H(y)
def predict_x(self,y,bs=256):
assert y.shape[-1] == self.y_dim
N = y.shape[0]
x_pred = np.zeros(shape=(N, self.x_dim+self.nb_classes))
x_onehot = np.zeros(shape=(N, self.nb_classes))
for b in range(int(np.ceil(N*1.0 / bs))):
if (b+1)*bs > N:
ind = np.arange(b*bs, N)
else:
ind = np.arange(b*bs, (b+1)*bs)
batch_y = y[ind, :]
batch_x_,batch_x_onehot_ = self.sess.run([self.x_latent_, self.x_onehot_], feed_dict={self.y:batch_y})
x_pred[ind, :] = batch_x_
x_onehot[ind, :] = batch_x_onehot_
return x_pred, x_onehot
def save(self,batch_idx):
checkpoint_dir = 'checkpoint/{}_{}_x_dim={}_y_dim={}_alpha={}_beta={}'.format(self.timestamp,self.data,self.x_dim, self.y_dim, self.alpha, self.beta)
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
self.saver.save(self.sess, os.path.join(checkpoint_dir, 'model.ckpt'),global_step=batch_idx)
def load(self, pre_trained = False, timestamp='',batch_idx=999):
if pre_trained == True:
print('Loading Pre-trained Model...')
checkpoint_dir = 'pre_trained_models/{}'.format(self.data)
self.saver.restore(self.sess, os.path.join(checkpoint_dir, 'model.ckpt-best'))
else:
if timestamp == '':
print('Best Timestamp not provided.')
checkpoint_dir = ''
else:
checkpoint_dir = 'checkpoint/{}_{}_x_dim={}_y_dim={}_alpha={}_beta={}'.format(timestamp,self.data,self.x_dim, self.y_dim, self.alpha, self.beta)
self.saver.restore(self.sess, os.path.join(checkpoint_dir, 'model.ckpt-%d'%batch_idx))
print('Restored model weights.')
if __name__ == '__main__':
parser = argparse.ArgumentParser('')
parser.add_argument('--data', type=str, default='Splenocyte',help='name of dataset')
parser.add_argument('--model', type=str, default='model',help='model definition')
parser.add_argument('--K', type=int, default=11,help='number of clusters')
parser.add_argument('--dx', type=int, default=10,help='dimension of Gaussian distribution')
parser.add_argument('--dy', type=int, default=20,help='dimension of preprocessed data')
parser.add_argument('--bs', type=int, default=64,help='batch size')
parser.add_argument('--nb_batches', type=int, default=50000,help='total number of training batches or the batch idx for loading pretrain model')
parser.add_argument('--alpha', type=float, default=10.0,help='coefficient of loss term')
parser.add_argument('--beta', type=float, default=10.0,help='coefficient of loss term')
parser.add_argument('--ratio', type=float, default=0.2,help='parameter in updating Caltegory distribution')
parser.add_argument('--low', type=float, default=0.03,help='low ratio for filtering peaks')
parser.add_argument('--timestamp', type=str, default='')
parser.add_argument('--train', type=bool, default=False,help='whether train from scratch')
parser.add_argument('--no_label', action='store_true',help='whether the dataset has label')
parser.add_argument('--mode', type=int, default=1,help='mode for 10x paired data')
args = parser.parse_args()
data = args.data
model = importlib.import_module(args.model)
nb_classes = args.K
x_dim = args.dx
y_dim = args.dy
batch_size = args.bs
nb_batches = args.nb_batches
alpha = args.alpha
beta = args.beta
ratio = args.ratio
low = args.low
timestamp = args.timestamp
is_train = args.train
has_label = not args.no_label
g_net = model.Generator(input_dim=x_dim,output_dim = y_dim,name='g_net',nb_layers=10,nb_units=512,concat_every_fcl=False)
h_net = model.Encoder(input_dim=y_dim,output_dim = x_dim+nb_classes,feat_dim=x_dim,name='h_net',nb_layers=10,nb_units=256)
dx_net = model.Discriminator(input_dim=x_dim,name='dx_net',nb_layers=2,nb_units=256)
dy_net = model.Discriminator(input_dim=y_dim,name='dy_net',nb_layers=2,nb_units=256)
pool = util.DataPool(10)
xs = util.Mixture_sampler(nb_classes=nb_classes,N=10000,dim=x_dim,sd=1)
if data == "PBMC10k":
mode = args.mode
ys = util.ARC_Sampler(name=data,n_components=int(y_dim/2),mode=mode)
else:
ys = util.scATAC_Sampler(data,y_dim,low,has_label)
model = scDEC(g_net, h_net, dx_net, dy_net, xs, ys, nb_classes, data, pool, batch_size, alpha, beta, is_train)
if args.train:
model.train(nb_batches=nb_batches)
else:
print('Attempting to Restore Model ...')
if timestamp == '':
model.load(pre_trained=True)
timestamp = 'pre-trained'
else:
model.load(pre_trained=False, timestamp = timestamp, batch_idx = nb_batches-1)
model.evaluate(timestamp,nb_batches-1)