-
Notifications
You must be signed in to change notification settings - Fork 27
/
Copy pathtest.py
148 lines (129 loc) · 5.22 KB
/
test.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
import os, math, time, argparse
os.environ["CUDA_VISIBLE_DEVICES"]="-1"
import numpy as np
import tensorflow as tf
from scipy.io import savemat
sess = tf.InteractiveSession()
from model.data.wrappers import *
########### setup params, data, etc ###########
# read params
parser = argparse.ArgumentParser(description='ODE2VAE test arguments')
parser.add_argument('--data_root', type=str, default='../neurips19/data',
help='root of the data folder')
parser.add_argument('--task', type=str, default='mocap_many',
help='experiment to execute')
parser.add_argument('--ckpt', type=str, default='checkpoints',
help='checkpoints file')
parser.add_argument('--subject_id', type=int, default=0,\
help='subject ID in mocap_single experiments')
opt = parser.parse_args()
opt = vars(opt)
for key in opt.keys():
print('{:s}: {:s}'.format(key, str(opt[key])))
locals().update(opt)
if not os.path.exists(os.path.join('plots', task)):
os.makedirs(os.path.join('plots', task))
ext = 'test'
dataset, N, T, D = load_data(data_root, task, subject_id=subject_id, plot=True)
saver = tf.train.import_meta_graph('{:s}.meta'.format(ckpt))
saver.restore(sess,'{:s}'.format(ckpt))
numpars = np.sum([np.prod(v.get_shape().as_list()) for v in tf.trainable_variables()])
print('{:d} trainable parameters'.format(numpars))
all_var_list = tf.trainable_variables()
for i in all_var_list:
print(i)
# tf.global_variables_initializer().run()
tf.tables_initializer().run()
graph = tf.get_default_graph()
dt = 0.1
x = graph.get_tensor_by_name("X:0")
t = graph.get_tensor_by_name("t:0")
is_train = graph.get_tensor_by_name("is_train:0")
D = int(sess.run(graph.get_tensor_by_name("D:0")))
q = int(sess.run(graph.get_tensor_by_name("q:0")))
L = int(sess.run(graph.get_tensor_by_name("L:0")))
amort_len = int(sess.run(graph.get_tensor_by_name("amort_len:0")))
fs_op = graph.get_tensor_by_name("f/S:0")
Tss = graph.get_tensor_by_name("Tss:0")
vt_op = graph.get_tensor_by_name("vt_L:0") # T,N,L,q # "transpose_2:0"
st_op = graph.get_tensor_by_name("st_L:0") # "transpose_3:0"
s0_log_sig_op = graph.get_tensor_by_name("s0_log_sigma_sq:0")
v0_log_sig_op = graph.get_tensor_by_name("v0_log_sigma_sq:0")
Xrec_op = graph.get_tensor_by_name("x_rec_mu_mu:0") # T*N,D # rec_mean:0
Xrec_op_L = graph.get_tensor_by_name("x_rec_mu_L:0") # T*N*L,D
def integrate(X,L=5,ts=None,fix_BNN=False,fix_enc=False):
if ts is None:
ts = dt * np.arange(0,X.shape[1],dtype=np.float32)
feed_dict = {x:X, t:ts, is_train:False,Tss:len(ts)}
if fix_BNN:
feed_dict[fs_op] = np.array([-20.0],dtype=np.float32)
if fix_enc:
zero = -20*np.ones([X.shape[0],3])
feed_dict[s0_log_sig_op] = zero
feed_dict[v0_log_sig_op] = zero
sts = []
for l in range(L):
st = sess.run(st_op, feed_dict=feed_dict)
if 'nonuniform' in task:
st = np.expand_dims(st,2)
sts.append(st)
st = np.concatenate(sts,2)
return st # T,N,L,q
def reconstruct(X,L=5,ts=None,fix_BNN=False,fix_enc=False):
if ts is None:
ts = dt * np.arange(0,X.shape[1],dtype=np.float32)
feed_dict = {x:X, t:ts, is_train:False}
if fix_BNN:
feed_dict[fs_op] = np.array([-20.0],dtype=np.float32)
if fix_enc:
zero = -20*np.ones([X.shape[0],3])
feed_dict[s0_log_sig_op] = zero
feed_dict[v0_log_sig_op] = zero
Xrecs = []
for l in range(L):
# Xrec = sess.run(Xrec_op, feed_dict={x: X, t:ts, is_train:False, Tss:X.shape[1]})
Xrec = sess.run(Xrec_op_L, feed_dict={x: X, t:ts, is_train:False, Tss:X.shape[1]})
Xrec = np.reshape(Xrec,(-1,X.shape[0],D)) # [T,N,D]
Xrec = np.transpose(Xrec,[1,0,2]) # [N,T,D]
Xrec = np.expand_dims(Xrec,1)
Xrecs.append(Xrec)
Xrec = np.concatenate(Xrecs,1) # [N,L,T,D]
return Xrec
L = 5
X = dataset.test.next_batch(5)[0]
ts = dt * np.arange(0,X.shape[1],dtype=np.float32)
if 'nonuniform' in task:
ts = np.tile(ts,[X.shape[0],1])
zt = integrate(X,L=L,ts=ts) # sampled latent trajectories
if q==2 or q==3:
fname='plots/{:s}/latent_{:s}.png'.format(task,ext)
cols = ['b','r','g','m','c','k','y']
[T,N,L,q] = zt.shape
zt = np.transpose(zt,[1,2,3,0]) # N,L,q,T
plt.figure(1,(8,8))
if q==3:
from mpl_toolkits.mplot3d import axes3d
ax = plt.subplot(111,projection='3d')
for j in range(N):
for i in range(L):
if q==2:
ax.plot(zt[j,i,0,:],zt[j,i,1,:],'-',color=cols[i])
ax.scatter(zt[j,i,0,0],zt[j,i,1,0],'o',color='k',linewidth=1)
else:
ax.plot(zt[j,i,0,:],zt[j,i,1,:],zt[j,i,2,:],'-',color=cols[i])
ax.scatter(zt[j,i,0,0],zt[j,i,1,0],zt[j,i,2,0],'o',color='k',linewidth=1)
plt.savefig(fname)
plt.close()
X,ts = dataset.train.next_batch(5)
if 'nonuniform' not in task:
ts = ts[0]
Xrec = reconstruct(X,L=1,ts=ts) # samples from the decoder
Xrec = np.squeeze(Xrec,1) # N,T,D
plot_reconstructions(task,X,Xrec,ts,show=False,fname='plots/{:s}/rec_{:s}.png'.format(task,'train'))
X = dataset.test.next_batch(5)[0]
ts = dt * np.arange(0,X.shape[1],dtype=np.int32)
if 'nonuniform' in task:
ts = np.tile(ts,[X.shape[0],1])
Xrec = reconstruct(X,L=1,ts=ts) # samples from the decoder
Xrec = np.squeeze(Xrec,1) # N,T,D
plot_reconstructions(task,X,Xrec,ts,show=False,fname='plots/{:s}/rec_{:s}.png'.format(task,ext))