def next_valid_batch(self, x_, y_, x_skeleton_):
if len(self.pos_valid) == 0: self.shuffle_valid()
pos = self.pos_valid.pop()
#print type(self.Mean_CNN)
#print type(self.Mean_CNN)
if not isinstance(self.Mean_CNN,int):
x_.set_value(normalize(self.x_valid[pos:pos+self.batch_size].astype(config.floatX), self.Mean_CNN, self.Std_CNN) , borrow=True)
else:
x_.set_value(normalize(self.x_valid[pos:pos+self.batch_size], self.Mean_CNN, self.Std_CNN) , borrow=True)
y_.set_value(self.y_valid[pos:pos+self.batch_size] , borrow=True)
x_skeleton_.set_value( normalize(self.x_valid_skeleton_feature[pos:pos+self.batch_size], self.Mean1, self.Std1) , borrow=True)
video, Feature_gesture = sample.get_test_data_wudi_lio(used_joints)
assert video.shape[0] == Feature_gesture.shape[0]# -*- coding: utf-8 -*-
print "finish preprocessing"
out_file = open(save_path, 'wb')
cPickle.dump({"video":video, "Feature_gesture":Feature_gesture}, out_file, protocol=cPickle.HIGHEST_PROTOCOL)
out_file.close()
print "start computing likelihood"
observ_likelihood = numpy.empty(shape=(video.shape[0],20*STATE_NO+1)) # 20 classed * 5 states + 1 ergodic state
for batchnumber in xrange(video.shape[0]/batch.micro):
video_temp = video[batch.micro*batchnumber:batch.micro*(batchnumber+1),:]
skel_temp = Feature_gesture[batch.micro*batchnumber:batch.micro*(batchnumber+1),:]
x_.set_value(normalize(video_temp, Mean_CNN, Std_CNN).astype("float32"),borrow=True)
x_skeleton_.set_value(normalize(skel_temp,Mean_skel, Std_skel).astype("float32"), borrow=True)
observ_likelihood[batch.micro*batchnumber:batch.micro*(batchnumber+1),:] = p_y_given_x()
# because input batch number should be 64, so here it is a bit of hack:
video_temp_1 = video[batch.micro* (batchnumber+1):,:]
video_temp_2 = numpy.zeros(shape=(64-video_temp_1.shape[0], 2, 2, 4, 64, 64))
video_temp = numpy.concatenate((video_temp_1, video_temp_2), axis=0)
skel_temp_1 = Feature_gesture[batch.micro* (batchnumber+1):,:]
skel_temp_2 = numpy.zeros(shape=(64-skel_temp_1.shape[0],891))
skel_temp = numpy.concatenate((skel_temp_1, skel_temp_2), axis=0)
x_.set_value(normalize(video_temp, Mean_CNN, Std_CNN).astype("float32"),borrow=True)
x_skeleton_.set_value(normalize(skel_temp,Mean_skel, Std_skel).astype("float32"), borrow=True)
ob_temp = p_y_given_x()
"video": video,
"Feature_gesture": Feature_gesture
},
out_file,
protocol=cPickle.HIGHEST_PROTOCOL)
out_file.close()
print "start computing likelihood"
observ_likelihood = numpy.empty(
shape=(video.shape[0], 20 * STATE_NO +
1)) # 20 classed * 5 states + 1 ergodic state
for batchnumber in xrange(video.shape[0] / batch.micro):
skel_temp = Feature_gesture[batch.micro * batchnumber:batch.micro *
(batchnumber + 1), :]
x_skeleton_.set_value(normalize(skel_temp, Mean_skel,
Std_skel).astype("float32"),
borrow=True)
observ_likelihood[batch.micro * batchnumber:batch.micro *
(batchnumber + 1), :] = test_model()
# because input batch number should be 64, so here it is a bit of hack:
skel_temp_1 = Feature_gesture[batch.micro * (batchnumber + 1):, :]
skel_temp_2 = numpy.zeros(shape=(64 - skel_temp_1.shape[0], 891))
skel_temp = numpy.concatenate((skel_temp_1, skel_temp_2), axis=0)
x_skeleton_.set_value(normalize(skel_temp, Mean_skel,
Std_skel).astype("float32"),
borrow=True)
ob_temp = test_model()
observ_likelihood[batch.micro *
(batchnumber +
1):, :] = ob_temp[:video_temp_1.shape[0], :]
"Feature_gesture": Feature_gesture
},
out_file,
protocol=cPickle.HIGHEST_PROTOCOL)
out_file.close()
print "start computing likelihood"
observ_likelihood = numpy.empty(
shape=(video.shape[0], 20 * STATE_NO +
1)) # 20 classed * 5 states + 1 ergodic state
for batchnumber in xrange(video.shape[0] / batch.micro):
video_temp = video[batch.micro * batchnumber:batch.micro *
(batchnumber + 1), :]
x_.set_value(normalize(video_temp, Mean_CNN,
Std_CNN).astype("float32"),
borrow=True)
observ_likelihood[batch.micro * batchnumber:batch.micro *
(batchnumber + 1), :] = test_model()
# because input batch number should be 64, so here it is a bit of hack:
video_temp_1 = video[batch.micro * (batchnumber + 1):, :]
video_temp_2 = numpy.zeros(shape=(64 - video_temp_1.shape[0], 2, 2, 4,
64, 64))
video_temp = numpy.concatenate((video_temp_1, video_temp_2), axis=0)
x_.set_value(normalize(video_temp, Mean_CNN,
Std_CNN).astype("float32"),
borrow=True)
ob_temp = test_model()
def next_valid_batch(self, x_, y_, x_skeleton_):
if len(self.pos_valid) == 0: self.shuffle_valid()
pos = self.pos_valid.pop()
x_.set_value(normalize(self.x_valid[pos:pos+self.batch_size], self.Mean_CNN, self.Std_CNN) , borrow=True)
y_.set_value(self.y_valid[pos:pos+self.batch_size] , borrow=True)
x_skeleton_.set_value( normalize(self.x_valid_skeleton_feature[pos:pos+self.batch_size], self.Mean1, self.Std1) , borrow=True)
protocol=cPickle.HIGHEST_PROTOCOL)
out_file.close()
print "start computing likelihood"
observ_likelihood = numpy.empty(
shape=(video.shape[0], 20 * STATE_NO +
1)) # 20 classed * 5 states + 1 ergodic state
for batchnumber in xrange(video.shape[0] / batch.micro):
video_temp = video[batch.micro * batchnumber:batch.micro *
(batchnumber + 1), :]
skel_temp = Feature_gesture[batch.micro * batchnumber:batch.micro *
(batchnumber + 1), :]
if not isinstance(Mean_CNN, int):
x_.set_value(normalize(video_temp.astype("float32"), Mean_CNN,
Std_CNN).astype("float32"),
borrow=True)
else:
x_.set_value(normalize(video_temp, Mean_CNN,
Std_CNN).astype("float32"),
borrow=True)
if not isinstance(Mean_skel, int):
x_skeleton_.set_value(normalize(skel_temp.astype("float32"),
Mean_skel,
Std_skel).astype("float32"),
borrow=True)
else:
x_skeleton_.set_value(normalize(skel_temp, Mean_skel,
Std_skel).astype("float32"),
borrow=True)