def train(epoch):
model.train()
data = shuffle_data(train_data, 1)
total_loss = 0.0
for num, i in enumerate(range(0, len(data), args.batch_size)):
one = data[i:i + args.batch_size]
query, _ = padding([x[0] for x in one], max_len=50)
passage, _ = padding([x[1] for x in one], max_len=350)
answer = pad_answer([x[2] for x in one])
query, passage, answer = torch.LongTensor(query), torch.LongTensor(
passage), torch.LongTensor(answer)
if args.cuda:
query = query.cuda()
passage = passage.cuda()
answer = answer.cuda()
optimizer.zero_grad()
loss = model([query, passage, answer, True])
loss.backward()
total_loss += loss.item()
optimizer.step()
if (num + 1) % args.log_interval == 0:
print(
'|------epoch {:d} train error is {:f} eclipse {:.2f}%------|'
.format(epoch, total_loss / args.log_interval,
i * 100.0 / len(data)))
total_loss = 0
def inference():
model.eval()
predictions = []
with torch.no_grad():
for i in range(0, len(data), args.batch_size):
one = data[i:i + args.batch_size]
query, _ = padding([x[0] for x in one], max_len=50)
passage, _ = padding([x[1] for x in one], max_len=300)
answer = pad_answer([x[2] for x in one])
str_words = [x[-1] for x in one]
ids = [x[3] for x in one]
query, passage, answer = torch.LongTensor(query), torch.LongTensor(
passage), torch.LongTensor(answer)
if args.cuda:
query = query.cuda()
passage = passage.cuda()
answer = answer.cuda()
output = model([query, passage, answer, False])
for q_id, prediction, candidates in zip(ids, output, str_words):
prediction_answer = u''.join(candidates[prediction])
predictions.append(str(q_id) + '\t' + prediction_answer)
outputs = u'\n'.join(predictions)
with codecs.open(args.output, 'w', encoding='utf-8') as f:
f.write(outputs)
print('done!')
def __init__(self, filename, *args, **kwargs):
name = 'ffmpeg'
tags = '/hafarm/ffmpeg'
super(BatchMp4, self).__init__(name, tags, *args, **kwargs)
scene_file_path, _, _, _ = utils.padding(filename, 'nuke')
base, file = os.path.split(scene_file_path)
file, _ = os.path.splitext(file)
inputfile = os.path.join(base, const.PROXY_POSTFIX, file + '.jpg')
outputfile = os.path.join(base, utils.padding(filename)[0] + 'mp4')
self.parms['command_arg'] = ['-y -r 25 -i %s -an -vcodec libx264 -vpre slow -crf 26 -threads 1 %s' % (inputfile, outputfile)]
self.parms['exe'] = 'ffmpeg'
self.parms['job_name'] << { 'render_driver_type': 'mp4' }
def test():
vES = elasticsearchVisitor()
request_amount = 10
model.eval()
with torch.no_grad():
while True:
raw_post = raw_input('input:')
if raw_post == 'exit':
break
# post规范化
raw_post = ' '.join(word_tokenize(raw_post))
raw_post = ''.join([w.lower() for w in raw_post])
# 选出90条候选评论
candi_comms = vES.visitSpaceQuery(raw_post, request_amount * 3)
# comms规范化
candi_comms = filter_comms(candi_comms[::-1], request_amount * 3,
30)
post = map_sent_to_id(raw_post.split())
posts, _ = padding([post for _ in range(request_amount * 3)],
max_len=30)
posts = torch.LongTensor(posts)
if args.cuda:
posts = posts.cuda()
comms = list()
for comm in candi_comms:
comm = word_tokenize(comm)
comm = map_sent_to_id(comm)
comms.append(comm)
comms, _ = padding([comm for comm in comms], max_len=30)
comms = torch.LongTensor(comms)
if args.cuda:
comms = comms.cuda()
start = time.time()
output = model([posts, comms, False])
print("model cost time %f" % (time.time() - start))
print('all comm candidates:')
score_comm = list()
for i, comm in enumerate(candi_comms):
jaccard = jaccard_similarity(raw_post, comm)
score_comm.append([output[i, 0].item(), jaccard, comm])
score_comm = sorted(score_comm, key=lambda x: x[0], reverse=True)
for score, jaccard, comm in score_comm:
print(str(score) + '\t' + str(jaccard) + '\t' + comm)
print("------------------------------")
print('match comms:')
for score, jaccard, comm in score_comm:
if score > 0.9 and jaccard < 0.4:
print(comm)
def dataset_input_fn(is_train, batch_size=64, split=1):
sounds, labels = train[split - 1] if is_train is True else val[split - 1]
labels = np.array(labels).reshape((-1, 1))
dataset = tf.data.Dataset.from_generator(
lambda: zip(sounds, labels),
output_types=(tf.float32, tf.int32),
output_shapes=(tf.TensorShape([None]), tf.TensorShape(1)))
# if is_train:
# if opt.strongAugment:
# dataset = dataset.map(U.random_scale(1.25))
dataset = dataset.map(U.padding(opt.inputLength // 2))
dataset = dataset.map(U.random_crop(opt.inputLength))
dataset = dataset.map(U.normalize(float(2**16 / 2)))
dataset = dataset.shuffle(1000)
# else:
# # if not opt.longAudio:
# dataset = dataset.map(U.padding(opt.inputLength // 2))
# dataset = dataset.map(U.random_crop(opt.inputLength))
# dataset = dataset.map(U.normalize(float(2 ** 16 / 2)))
# # dataset = dataset.map(U.multi_crop(opt.inputLength, opt.nCrops))
dataset = dataset.batch(batch_size)
dataset = dataset.map(U.reshape([batch_size, -1, 1]))
iterator = dataset.make_one_shot_iterator()
return iterator.get_next()
def get_test_loader(full_data, support, query, pad_idx):
loader = []
for filename in full_data:
# support
support_data = full_data[filename]['neg']['support_data'][
0:support] + full_data[filename]['pos']['support_data'][0:support]
support_data = batch_padding(support_data, pad_idx)
support_target = full_data[filename]['neg']['support_target'][
0:support] + full_data[filename]['pos']['support_target'][0:support]
support_target = torch.tensor(support_target)
# query
neg_dl = DataLoader(Dataset(full_data[filename]['neg'], pad_idx),
batch_size=query * 2,
shuffle=False,
drop_last=False,
**kwargs)
pos_dl = DataLoader(Dataset(full_data[filename]['pos'], pad_idx),
batch_size=query * 2,
shuffle=False,
drop_last=False,
**kwargs)
# combine
for dl in [neg_dl, pos_dl]:
for batch_data, batch_target in dl:
support_data_cp, support_target_cp = copy.deepcopy(
support_data), copy.deepcopy(support_target)
support_data_cp, batch_data = padding(support_data_cp,
batch_data, pad_idx)
data = torch.cat([support_data_cp, batch_data], dim=0)
target = torch.cat([support_target_cp, batch_target], dim=0)
loader.append((data, target))
print('test loader length', len(loader))
return loader
def __init__(self, filename, *args, **kwargs):
"""
Args:
filename (str):
Kwargs:
start (int):
end (int):
"""
name = 'debug_images.py'
tags = '/hafarm/debug_images'
super(BatchDebug, self).__init__(name, tags, *args, **kwargs)
self.parms['start_frame'] = kwargs.get('start', 1)
self.parms['end_frame'] = kwargs.get('end', 1)
scene_file_path, _, frame_padding_length, ext = utils.padding(filename)
path, file = os.path.split(filename)
path = os.path.join(path, const.DEBUG_POSTFIX)
self.parms['pre_render_script'] = 'mkdir -p %s' % path
self.parms['scene_file'] << {
'scene_fullpath': scene_file_path + const.TASK_ID_PADDED + ext
}
self.parms['exe'] = '$HAFARM_HOME/scripts/debug_images.py'
self.parms['command_arg'] = [
'--job %s' % self.parms['job_name'], '--save_json -i'
]
self.parms['frame_padding_length'] = int(frame_padding_length)
self.parms['job_name'] << {'render_driver_type': 'debug'}
def read_iris_data(self, img_paths, margin=5, is_augment=False):
batch_imgs = np.zeros((len(img_paths), self.input_img_shape[1], self.input_img_shape[1], 1), dtype=np.float32)
batch_labels = np.zeros((len(img_paths), 1), dtype=np.uint8)
for i, img_path in enumerate(img_paths):
mask = np.zeros((self.img_shape[0], self.img_shape[1], 3), dtype=np.uint8)
# Extract Iris part
img_combine = cv2.imread(img_path)
img = img_combine[:, :self.img_shape[1], :]
seg = img_combine[:, self.img_shape[1]:, :]
if is_augment is True:
# Data augmentation: random brightness + random rotation
img_aug, seg_aug = utils.data_augmentation(img, seg)
mask[:, :, :][seg_aug[:, :, 1] == 204] = 1
img = img_aug * mask
else:
mask[:, :, :][seg[:, :, 1] == 204] = 1
img = img * mask
# Cropping iris part
x, y, w, h = cv2.boundingRect(mask[:, :, 1])
new_x = np.maximum(0, x - margin)
new_y = np.maximum(0, y - margin)
crop_img = img[new_y:new_y + h + margin, new_x:new_x + w + margin, 1] # Extract more bigger area
# Padding to the required size by preserving ratio of height and width
batch_imgs[i, :, :, 0] = utils.padding(crop_img)
batch_labels[i] = self.convert_to_cls(img_path)
return batch_imgs, batch_labels
def __init__(self, filename, *args, **kwargs):
"""
Args:
filename (str):
Kwargs:
resend_frames (bool): Current state to be in.
ifd_path
mad_threshold (float):
"""
name = 'generate_render_report.py'
tags = '/hafarm/generate_render_report'
super(BatchReportsMerger, self).__init__(name, tags, *args, **kwargs)
resend_frames = kwargs.get('resend_frames', False)
ifd_path = kwargs.get('ifd_path')
mad_threshold = kwargs.get('mad_threshold', 5.0)
send_email = '--send_email'
ifd_path = '--ifd_path %s' % ifd_path if ifd_path else ''
resend_frames = '--resend_frames' if resend_frames else ''
path, filename = os.path.split(filename)
scene_file_path, _, _, _ = utils.padding(filename, 'shell')
log_path = os.path.join(path, const.DEBUG_POSTFIX)
self.parms['job_name'] << {'render_driver_type': 'reports'}
self.parms['scene_file'] << {
'scene_file_path': log_path,
'scene_file_basename': scene_file_path,
'scene_file_ext': 'json'
}
self.parms['exe'] = '$HAFARM_HOME/scripts/generate_render_report.py'
self.parms['command_arg'] = [
send_email, ifd_path, resend_frames,
"--mad_threshold %s" % (mad_threshold), "--save_html"
]
def pastex25(img):
width, height = img.size
imgx9 = padding(img, (height * 5, width * 5), True)
angles = [
-15,
-30,
-45,
-60,
-75,
-90,
-105,
-120,
-135,
-150,
-165,
-180,
15,
30,
45,
60,
75,
90,
105,
120,
135,
150,
165,
180,
]
xys = [
[0, 0],
[width, 0],
[width * 2, 0],
[width * 3, 0],
[width * 4, 0],
[0, height],
[width, height],
[width * 2, height],
[width * 3, height],
[width * 4, height],
[0, height * 2],
[width, height * 2],
[width * 3, height * 2],
[width * 4, height * 2],
[0, height * 3],
[width, height * 3],
[width * 2, height * 3],
[width * 3, height * 3],
[width * 4, height * 3],
[0, height * 4],
[width, height * 4],
[width * 2, height * 4],
[width * 3, height * 4],
[width * 4, height * 4],
]
for angle, (x, y) in zip(angles, xys):
aimg = img.rotate(angle, resample=Image.BICUBIC)
imgx9.paste(aimg, (x, y, x + width, y + height))
return imgx9
def gen_file(self, data_set, size=True):
if size:
size = self.config.batch_size
else:
size = 5 * self.config.batch_size
samples = random.sample(list(data_set), size)
inputs, length, labels = zip(*self.load_file(samples))
max_len = max(length)
inputs = padding(inputs, max_len)
return inputs, length, labels
def test():
model.eval()
r, a = 0.0, 0.0
with torch.no_grad():
for i in range(0, len(dev_data), args.batch_size):
one = dev_data[i:i + args.batch_size]
query, _ = padding([x[0] for x in one], max_len=50)
passage, _ = padding([x[1] for x in one], max_len=500)
answer = pad_answer([x[2] for x in one])
query, passage, answer = torch.LongTensor(query), torch.LongTensor(
passage), torch.LongTensor(answer)
if args.cuda:
query = query.cuda()
passage = passage.cuda()
answer = answer.cuda()
output = model([query, passage, answer, False])
r += torch.eq(output, 0).sum().item()
a += len(one)
return r * 100.0 / a
def preprocess_setup(self):
if self.train:
funcs = []
if self.opt.strongAugment:
funcs += [U.random_scale(1.25)]
funcs += [
U.padding(self.opt.inputLength // 2),
U.random_crop(self.opt.inputLength),
U.normalize(32768.0),
]
else:
funcs = [
U.padding(self.opt.inputLength // 2),
U.normalize(32768.0),
U.multi_crop(self.opt.inputLength, self.opt.nCrops),
]
return funcs
def handle_human_parsing(self, req):
if self.net == None:
try:
self.net = caffe.Net(self.model_path, self.weight_path,
caffe.TEST)
except:
rospy.logerr("Error, cannot load deep_net to the GPU")
self.net = None
self.service_queue -= 1
return HumanParsingResponse()
try:
img = self.br.imgmsg_to_cv2(req.rgb_img, desired_encoding="bgr8")
if self.gpu_id >= 0:
caffe.set_mode_gpu()
caffe.set_device(self.gpu_id)
else:
caffe.set_mode_cpu()
cv2.imwrite(
os.path.dirname(os.path.realpath(__file__)) +
"/../images/rgb_image.png", img)
img_padding = padding(img, 1)
img_padding = img_padding.astype(np.float32)
img_padding -= self.mean
data = img_padding.transpose((2, 0, 1))
self.net.blobs['data'].data[0, ...] = data
out = self.net.forward()
prediction = self.net.blobs['fc8_mask'].data[0, ...][0]
prediction = prediction[1:481, 1:641]
prediction = prediction.astype(np.int8)
prediction = cv2.merge((prediction, prediction, prediction))
prediction_rgb = np.zeros(prediction.shape, dtype=np.uint8)
label_colours_bgr = self.label_colours[..., ::-1]
cv2.LUT(prediction, label_colours_bgr, prediction_rgb)
cv2.imwrite(
os.path.dirname(os.path.realpath(__file__)) +
"/../images/prediction_rgb.png", prediction_rgb)
overlay_img = overlay(img, prediction_rgb)
cv2.imwrite(
os.path.dirname(os.path.realpath(__file__)) +
"/../images/overlay.png", overlay_img)
segmentation_img_msg = self.br.cv2_to_imgmsg(prediction_rgb,
encoding="bgr8")
self.net = None
return HumanParsingResponse(segmentation_img=segmentation_img_msg)
except cv_bridge.CvBridgeError as e:
rospy.logerr("CVBridge exception %s", e)
return HumanParsingResponse()
def combine_batch(self, neg_data, neg_target, pos_data, pos_target):
neg_data, pos_data = padding(neg_data, pos_data, pad_idx=self.pad_idx)
# combine support data and query data
support_data = torch.cat([neg_data[0:self.support], pos_data[0:self.support]], dim=0)
query_data = torch.cat([neg_data[self.support:], pos_data[self.support:]], dim=0)
data = torch.cat([support_data, query_data], dim=0)
# combine support target and query target
support_target = torch.cat([neg_target[0:self.support], pos_target[0:self.support]], dim=0)
query_target = torch.cat([neg_target[self.support:], pos_target[self.support:]], dim=0)
target = torch.cat([support_target, query_target], dim=0)
return data, target
def preprocess_setup(self):
if self.train:
funcs = [U.normalize(self.mean, self.std),
U.horizontal_flip(),
U.padding(4),
U.random_crop(32),
]
else:
funcs = [U.normalize(self.mean, self.std)]
return funcs
def load_train_data(self, positive_file, negative_file):
# Load data
positive_examples = []
negative_examples = []
with open(positive_file) as fin:
for line in fin:
line = line.strip()
line = line.split()
parse_line = [int(x) for x in line]
parse_line = padding(parse_line, self.seq_length, self.unk)
positive_examples.append(parse_line)
with open(negative_file) as fin:
for line in fin:
line = line.strip()
line = line.split()
parse_line = [int(x) for x in line]
parse_line = padding(parse_line, self.seq_length, self.unk)
negative_examples.append(parse_line)
num_negative = (
(len(positive_examples) + len(negative_examples)) //
self.batch_size) * self.batch_size - len(positive_examples)
negative_examples = negative_examples[:num_negative]
self.sentences = np.array(positive_examples + negative_examples)
# Generate labels
positive_labels = [[0, 1] for _ in positive_examples]
negative_labels = [[1, 0] for _ in negative_examples]
self.labels = np.concatenate([positive_labels, negative_labels], 0)
# Shuffle the data
shuffle_indices = np.random.permutation(np.arange(len(self.labels)))
self.sentences = self.sentences[shuffle_indices]
self.labels = self.labels[shuffle_indices]
# Split batches
self.num_batch = int(len(self.labels) / self.batch_size)
#self.sentences = self.sentences[:self.num_batch * self.batch_size]
#self.labels = self.labels[:self.num_batch * self.batch_size]
self.sentences_batches = np.split(self.sentences, self.num_batch, 0)
self.labels_batches = np.split(self.labels, self.num_batch, 0)
self.pointer = 0
def get_ifd_files(ifds):
ifds = ifds.strip()
if not os.path.exists(ifds):
print >> sys.stderr, ('Error! Ifd file not found: "%s"'%ifds)
return ([],'','')
# Rediscover ifds:
# FIXME: should be simple unexpandedString()
seq_details = utils.padding(ifds)
# Find real file sequence on disk. Param could have $F4...
real_ifds = glob.glob(seq_details[0] + "*" + seq_details[-1])
real_ifds.sort()
if real_ifds == []:
print "Can't find ifds files: %s" % ifds
return real_ifds, os.path.split(seq_details[0])[1], seq_details[0] + const.TASK_ID + '.ifd'
def draw(game):
# 渲染函数
def _render(game, maps):
print(game.canvas['title'])
print(game.canvas['score'])
for i in range(len(maps)):
print maps[i]
print(game.canvas['tips'])
# 生成图像
maps = []
maps.append(game.canvas['splitor'])
for i in range(game.level):
row = '|'
for j in range(game.level):
if game.layout[i][j] == 0:
row += padding(' ', game.size) + '|'
else:
row += padding(str(game.layout[i][j]), game.size) + '|'
maps.append(row)
maps.append(game.canvas['splitor'])
# 渲染图像
_render(game, maps)
def create_cond_batches(self, data_file):
self.cond_stream = []
with open(data_file, 'r') as f:
for line in f:
line = line.strip()
line = line.split()
parse_line = [int(x) for x in line]
self.cond_stream.append(
padding(parse_line, self.cond_length, self.unk))
self.num_batch = int(len(self.cond_stream) / self.batch_size)
self.cond_stream = self.cond_stream[:self.num_batch * self.batch_size]
self.cond_batch = np.split(np.array(self.cond_stream), self.num_batch,
0)
self.cond_pointer = 0
def preprocess_setup(self):
if self.opt.plus:
normalize = U.zero_mean
else:
normalize = U.normalize
if self.train and self.opt.noDataAug != True:
funcs = [normalize(self.mean, self.std),
U.horizontal_flip(),
U.padding(4),
U.random_crop(32),
]
else:
funcs = [normalize(self.mean, self.std)]
return funcs
def load_cond(self, input_file, cond_length, unk):
"""Get a list of unique conditions"""
f = codecs.open(input_file, 'r', 'utf-8')
conditions = []
lines = f.readlines()
if lines[-1].strip() == '':
print("deleted the last element:", lines[-1])
lines = lines[:-1]
lines = list(set(lines))
for line in lines:
line = map(int, line.strip().split())
line = padding(line, cond_length, unk)
if not line in conditions:
conditions.append(line)
self.cond = np.array(conditions)
self.n_cond = len(conditions)
def forward(self, xs, train):
"""
xs: [[int=ID]]
ys: T x (N,T)
"""
xs, ms = utils.padding(xs, head=True, with_mask=True) # (N, T), (N, T)
xs = utils.convert_ndarray_to_variable(xs, seq=True,
train=train) # T x (N,)
ms = utils.convert_ndarray_to_variable(ms, seq=True,
train=train) # T x (N,)
es = self.embed_words(xs, train=train)
g = self.aggregate(es, ms)
# Note: Here, we assume that "es" follows the original order
ys, outputs = self.reorder(es, g, ms, train=train) # bottleneck
return ys, outputs
def __getitem__(self, idx):
label = self.Label_list[idx]
if self.stage == 'train' and not self.whole:
data = np.load(self.Data_dir + self.Data_list[idx] + '.npy',
mmap_mode='r').astype(np.float32)
patch = self.patch_sampler.random_sample(data)
if self.transform:
patch = self.transform.apply(patch).astype(np.float32)
patch = np.expand_dims(patch, axis=0)
return patch, label
else:
data = np.load(self.Data_dir + self.Data_list[idx] +
'.npy').astype(np.float32)
data = np.expand_dims(padding(data, win_size=self.patch_size // 2),
axis=0)
return data, label
def get_haiku_conddis(kigo):
with open(path_to_token2id) as f:
token2id = pickle.load(f)
vocab_size = len(token2id)
UNK = token2id.get('<UNK>', 0)
generator = Generator(vocab_size,
BATCH_SIZE,
EMB_DIM,
HIDDEN_DIM,
SEQ_LENGTH,
COND_LENGTH,
START_TOKEN,
is_cond=1)
discriminator = Discriminator(sequence_length=SEQ_LENGTH,
cond_length=COND_LENGTH,
num_classes=2,
vocab_size=vocab_size,
batch_size=BATCH_SIZE,
embedding_size=dis_embedding_dim,
filter_sizes=dis_filter_sizes,
num_filters=dis_num_filters,
l2_reg_lambda=dis_l2_reg_lambda,
is_cond=1)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
saver.restore(sess, path_to_generator)
if isinstance(kigo, str):
kigo = kigo.decode('utf-8')
cond = map(lambda x: token2id.get(x, 0), [kigo])
cond = np.array(padding(cond, COND_LENGTH, UNK) * BATCH_SIZE).reshape(
BATCH_SIZE, COND_LENGTH)
generated_sequences = generator.generate(sess, cond=cond)
sess.close()
id2token = {k: v for v, k in token2id.items()}
generated_haikus = map(
lambda y: map(lambda x: id2token.get(x, '<UNK>'), y),
generated_sequences)
generated_haikus = map(lambda x: re.sub(r' <UNK>', '', ' '.join(x)),
generated_haikus)
return generated_haikus
def gen_file(self, data_set, size=True):
if size:
size = self.config.batch_size
samples = random.sample(list(data_set), size)
else:
samples = data_set
p, q, a = zip(samples)
inputs = [q[i] + [0] + p[i] for i in range(size)]
length = [len(it) for it in inputs]
max_len = max(length)
inputs = padding(inputs, max_len)
for row in inputs:
for col in row:
if col < 0 or col >= Config.vocab_size:
print("BUG!!!!!! index: {}".format(col))
labels = np.array(a).flatten()
length = np.array(length)
return inputs, length, labels
def pastex9(img):
width, height = img.size
imgx9 = padding(img, (height * 3, width * 3), True)
angles = [-30, -60, -90, 90, 30, 60, 45, -45]
xys = [
[0, 0],
[width, 0],
[width * 2, 0],
[0, height],
[width * 2, height],
[0, height * 2],
[width, height * 2],
[width * 2, height * 2],
]
for angle, (x, y) in zip(angles, xys):
aimg = img.rotate(angle, resample=Image.BICUBIC)
imgx9.paste(aimg, (x, y, x + width, y + height))
return imgx9
def func(flag, maxlen=maxlen):
if flag == 'lens':
lens = np.array(map(len, self.AccGyo[cur_id]))
lens = [get_closest_maxlen(L) for L in lens]
return lens
elif flag == 'AccGyo' or flag == 'XYZ':
res = self.__dict__[flag][cur_id]
maxlen = maxlen or max(map(len, res))
maxlen = get_closest_maxlen(maxlen)
res = padding(res, maxlen)
if self.time_major:
res = np.transpose(res, (1, 0, 2))
if self.expand_dim:
res = np.expand_dims(res, -1)
return res
elif flag == 'labels':
return self.labels[cur_id]
elif flag == 'filenames':
return self.fs[cur_id]
def gen_file(self):
samples = random.sample(list(self.data_set), 1)
gen_tuples = questions_to_token_ids(samples, self.vocabulary)
print(gen_tuples)
print("shape: {}".format(np.asarray(gen_tuples).shape))
inp = input("==========\n")
p, q, a = zip(*gen_tuples)
print("p: {}\tq: {}\ta: {}".format(
np.asarray(p).shape,
np.asarray(q).shape,
np.asarray(a).shape))
inp = input("==========\n")
inputs = np.concatenate((p, np.zeros(
(Config.batch_size, 1), dtype=int), q),
axis=1)
length = [len(it) for it in inputs]
max_len = max(length)
inputs = [padding(seq, max_len) for seq in inputs]
return inputs, length, a
def forward(
model, batch_sents, batch_labels,
lmd, identity_penalty,
train):
ys, _ = model.forward(batch_sents, train=train) # T x (N,T)
ys = F.concat(ys, axis=0) # => (T*N, T)
ts, M = utils.padding(batch_labels, head=True, with_mask=True) # => (N, T), (N, T)
ts = ts.T # => (T, N)
ts = ts.reshape(-1,) # => (T*N,)
M = M[:,None,:] * M[:,:,None] # => (N, T, T)
ts = utils.convert_ndarray_to_variable(ts, seq=False, train=train) # => (T*N,)
M = utils.convert_ndarray_to_variable(M, seq=False, train=train) # => (N, T, T)
loss = F.softmax_cross_entropy(ys, ts)
acc = F.accuracy(ys, ts, ignore_label=-1)
if identity_penalty:
loss_id = loss_identity_penalty(ys, M, train=train)
loss = loss + lmd * loss_id
return loss, acc
from layers import LSTM, FullConnected, TimeDistributed from models import Decoder, Encoder, Seq2seq, Sequential from utils import masking, padding rng.seed(123) # Preprocess data x1 = [2, 1, 1, 1, 2, 4, 2] x2 = [2, 1] x3 = [2, 1, 4, 3, 1] batch_value = np.asarray([x1, x2, x3]) vocab_size = 5 embedding_size = 4 encoder_hidden_size = 6 encoder = Encoder(vocab_size + 1, embedding_size, encoder_hidden_size) mask_value = masking(batch_value) padded_batch_value = padding(batch_value, 0) mask = shared(mask_value, name='mask') padded_batch = shared(padded_batch_value, name='padded_batch') H, C = encoder.forward(padded_batch, mask) (h1, c1) = encoder.forward2(x1) (h2, c2) = encoder.forward2(x2) (h3, c3) = encoder.forward2(x3) print(T.isclose(H, T.as_tensor_variable([h1, h2, h3])).eval()) print(T.isclose(C, T.as_tensor_variable([c1, c2, c3])).eval())
def begin_train(self):
N_EPOCHS = 30
N_BATCH = 5
N_TRAIN_INS = len(self.train_ending)
best_val_accuracy = 0
best_test_accuracy = 0
for epoch in range(N_EPOCHS):
print "epoch ", epoch,":"
shuffled_index_list = utils.shuffle_index(N_TRAIN_INS)
max_batch = N_TRAIN_INS/N_BATCH
start_time = time.time()
batch_count = 0
for batch in range(max_batch):
batch_index_list = [shuffled_index_list[i] for i in range(batch * N_BATCH, (batch+1) * N_BATCH)]
train_story = [self.train_story[index] for index in batch_index_list]
train_ending = [self.train_ending[index] for index in batch_index_list]
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
while np.any((np.asarray(batch_index_list) - neg_end_index_list) == 0):
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
neg_end1 = [self.train_ending[index] for index in neg_end_index_list]
train_story_matrix = utils.padding(train_story)
train_ending_matrix = utils.padding(train_ending)
neg_ending1_matrix = utils.padding(neg_end1)
train_story_mask = utils.mask_generator(train_story)
train_ending_mask = utils.mask_generator(train_ending)
neg_ending1_mask = utils.mask_generator(neg_end1)
self.train_func(train_story_matrix, train_story_mask,
train_ending_matrix, train_ending_mask,
neg_ending1_matrix, neg_ending1_mask)
batch_count += 1
if batch_count != 0 and batch_count % 10 == 0:
print "--", time.time() - start_time,"seconds--for 10 batches"
start_time = time.time()
# peek on val set every 5000 instances(1000 batches)
if batch_count != 0 and batch_count % 2000 == 0:
print"test on valid set..."
val_result = self.val_set_test()
print "accuracy is: ", val_result*100, "%"
if val_result > best_val_accuracy:
print "new best! test on test set..."
best_val_accuracy = val_result
self.saving_model('val')
test_accuracy = self.test_set_test()
print "test set accuracy: ", test_accuracy, "%"
if test_accuracy > best_test_accuracy:
best_test_accuracy = test_accuracy
print "saving model..."
self.saving_model('test')
print "reload best model for testing on test set"
self.reload_model('val')
print "test on test set..."
test_result = self.test_set_test()
print "accuracy is: ", test_result * 100, "%"
def run_FUCOS(**kwargs):
training_data = kwargs.get('training_data')
validation_data = kwargs.get('validation_data')
batchsize = kwargs.get('batchsize')
TRAIN = kwargs.get('TRAIN', True)
run = kwargs.get('run')
config_sess = tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)
config_sess.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=config_sess)
#build the model
model = []
with tf.device('/gpu:2'):
x = tf.placeholder(tf.float32, (None, 135, 240, 3), 'input')
y_ = tf.placeholder(tf.float32, (None, 135, 240, 1), 'gt')
keep_prob = tf.placeholder(tf.float32, name='dropout_prob')
with tf.variable_scope('conv1'):
conv1 = layers.ConvolutionalLayer(x, [135, 240, 3], [3, 3, 3, 64])
model.append(conv1)
with tf.variable_scope('conv2'):
conv2 = layers.ConvolutionalLayer(conv1.output(), conv1.get_output_shape(), [3, 3, 64, 64], pool=True)
model.append(conv2)
with tf.variable_scope('conv3'):
conv3 = layers.ConvolutionalLayer(conv2.output(), conv2.get_output_shape(), [3, 3, 64, 128])
model.append(conv3)
with tf.variable_scope('conv4'):
conv4 = layers.ConvolutionalLayer(conv3.output(), conv3.get_output_shape(), [3, 3, 128, 128], pool=True)
model.append(conv4)
with tf.variable_scope('conv5'):
conv5 = layers.ConvolutionalLayer(conv4.output(), conv4.get_output_shape(), [3, 3, 128, 256])
model.append(conv5)
with tf.variable_scope('conv6'):
conv6 = layers.ConvolutionalLayer(conv5.output(), conv5.get_output_shape(), [3, 3, 256, 256])
model.append(conv6)
with tf.variable_scope('conv7'):
conv7 = layers.ConvolutionalLayer(conv6.output(), conv6.get_output_shape(), [3, 3, 256, 256], pool=True)
model.append(conv7)
with tf.variable_scope('conv8'):
conv8 = layers.ConvolutionalLayer(conv7.output(), conv7.get_output_shape(), [3, 3, 256, 512])
model.append(conv8)
with tf.variable_scope('conv9'):
conv9 = layers.ConvolutionalLayer(conv8.output(), conv8.get_output_shape(), [3, 3, 512, 512])
model.append(conv9)
with tf.variable_scope('conv10'):
conv10 = layers.ConvolutionalLayer(conv9.output(), conv9.get_output_shape(), [3, 3, 512, 512], pool=True)
model.append(conv10)
with tf.variable_scope('conv11'):
conv11 = layers.ConvolutionalLayer(conv10.output(), conv10.get_output_shape(), [3, 3, 512, 512])
model.append(conv11)
with tf.variable_scope('conv12'):
conv12 = layers.ConvolutionalLayer(conv11.output(), conv11.get_output_shape(), [3, 3, 512, 512])
model.append(conv12)
with tf.variable_scope('conv13'):
conv13 = layers.ConvolutionalLayer(conv12.output(), conv12.get_output_shape(), [3, 3, 512, 512], pool=True)
model.append(conv13)
with tf.variable_scope('conv14'):
conv14 = layers.ConvolutionalLayer(conv13.output(), conv13.get_output_shape(), [7, 7, 512, 4096], drop_out=True,
drop_out_prob=keep_prob)
model.append(conv14)
with tf.variable_scope('conv15'):
conv15 = layers.ConvolutionalLayer(conv14.output(), conv14.get_output_shape(), [1, 1, 4096, 4096], drop_out=True,
drop_out_prob=keep_prob)
model.append(conv15)
with tf.variable_scope('convtrans1'):
deconv1 = layers.ConvolutionalTransposeLayer(conv15.output(), [4, 4, 60, 4096], None)
model.append(deconv1)
with tf.variable_scope('conv16'):
conv16 = layers.ConvolutionalLayer(conv10.output(), conv10.get_output_shape(), [1, 1, 512, 60])
model.append(conv16)
conv16_output = conv16.output()
sum1 = conv16_output + tf.image.resize_images(deconv1.output(), (tf.shape(conv16_output)[1],
tf.shape(conv16_output)[2]))
with tf.variable_scope('convtrans2'):
deconv2 = layers.ConvolutionalTransposeLayer(sum1, [4, 4, 60, 60], None)
model.append(deconv2)
with tf.variable_scope('conv17'):
conv17 = layers.ConvolutionalLayer(conv7.output(), conv7.get_output_shape(), [1, 1, 256, 60])
model.append(conv17)
conv17_output = conv17.output()
sum2 = conv17_output + tf.image.resize_images(deconv2.output(), (tf.shape(conv17_output)[1],
tf.shape(conv17_output)[2]))
with tf.variable_scope('convtrans3'):
deconv3 = layers.ConvolutionalTransposeLayer(sum2, [16, 16, 60, 60], None, deconv_stride=(1, 8, 8, 1))
model.append(deconv3)
with tf.variable_scope('conv18'):
conv18 = layers.ConvolutionalLayer(deconv3.output(), deconv3.get_output_shape(), [1, 1, 60, 12])
model.append(conv18)
with tf.variable_scope('conv19'):
conv19 = layers.ConvolutionalLayer(conv18.output(), conv18.get_output_shape_tensor(), [1, 1, 12, 1],
activation=function['linear'])
model.append(conv19)
y_pre_activation = tf.image.resize_images(conv19.output(), (135, 240)) #resize to match the ground truth's shape
y_pred = function['sigmoid'](y_pre_activation) #activate the output by sigmoid
cost = metrics.MultinoulliCrossEntropy(y_pre_activation, y_) #use binary cross entropy
var_list = tf.get_collection(tf.GraphKeys().TRAINABLE_VARIABLES)
L2 = sum([tf.reduce_mean(tf.square(theta)) #L2 regularization
for theta in (weight for weight in var_list if 'weights' in weight.name)])
cost += 1e-4 * L2
opt = tf.train.AdamOptimizer(1e-3, 0.9, 0.99, 1e-8).minimize(cost, var_list=var_list) #ADAM optimization
accuracy = tf.reduce_mean(tf.cast(tf.equal(tf.cast(y_pred >= 0.5, tf.uint8), tf.cast(y_, tf.uint8)), tf.float32))
saver = tf.train.Saver()
if TRAIN:
tf.Operation.run(tf.global_variables_initializer())
print('Loading VGG16 weights...')
load_weights('pretrained/vgg16_weights.npz', model, sess) #load pretrained VGG16 weights
best_valid_accuracy = 0.
best_valid_loss = np.inf
best_epoch = 0
epoch = 0
vote_to_terminate = 0
done_looping = False
print('TRAINING...')
start_training_time = time.time()
while epoch < 200 and not done_looping:
epoch += 1
num_iter_training = int(training_data[0].shape[0] / batchsize)
losses_train = 0.
accuracies_train = 0.
start_batch_time = time.time()
print('Epoch %d...' % epoch)
batch = next_batch(training_data, batchsize) #training
for b in batch:
fd = {x: b[0], y_: b[1], keep_prob: 0.3}
_, a, l = sess.run([opt, accuracy, cost], feed_dict=fd)
assert not np.isnan(l), 'Train failed with loss being NaN'
losses_train += l
accuracies_train += a
print('\ttraining loss: %s' % (losses_train / num_iter_training))
print('\ttraining accuracy: %s' % (accuracies_train / num_iter_training))
print('\tepoch %d took %.2f hours' % (epoch, (time.time() - start_batch_time) / 3600.))
num_iter_valid = int(validation_data[0].shape[0] / batchsize)
losses_valid = 0.
accuracies_valid = 0.
start_valid_time = time.time()
batch = next_batch(validation_data, batchsize) #validation
for b in batch:
fd = {x: b[0], y_: b[1], keep_prob: 1}
l, a = sess.run([cost, accuracy], feed_dict=fd)
losses_valid += l
accuracies_valid += a
avr_acc_valid = accuracies_valid / num_iter_valid
losses_valid /= num_iter_valid
print('\tvalidation took %.2f hours' % ((time.time() - start_valid_time) / 3600.))
print('\tvalidation loss: %s' % losses_valid)
print('\tvalidation accuracy: %s' % avr_acc_valid)
if losses_valid < best_valid_loss:
best_valid_loss = losses_valid
best_epoch = epoch
vote_to_terminate = 0
print('\tbest validation loss achieved: %.4f' % best_valid_loss)
save_path = saver.save(sess, run)
print("\tmodel saved in file: %s" % save_path)
else:
vote_to_terminate += 1
if vote_to_terminate > 30:
done_looping = True
print('Training ends after %.2f hours' % ((time.time() - start_training_time) / 3600.))
print('\tbest validation accuracy: %.2f' % best_valid_accuracy)
print('Training the model using all data available...')
total_training_data = (np.concatenate((training_data[0], validation_data[0])),
np.concatenate((training_data[1], validation_data[1])))
for i in range(best_epoch):
num_iter_training = int(total_training_data[0].shape[0] / batchsize)
losses_train = 0.
start_batch_time = time.time()
print('Epoch %d...' % (i+1))
batch = next_batch(total_training_data, batchsize) #training
for b in batch:
fd = {x: b[0], y_: b[1], keep_prob: 0.1}
_, _, l = sess.run([opt, accuracy, cost], feed_dict=fd)
assert not np.isnan(l), 'Train failed with loss being NaN'
losses_train += l
print('\ttraining loss: %s' % (losses_train / num_iter_training))
print('\tepoch %d took %.2f hours' % (i+1, (time.time() - start_batch_time) / 3600.))
else: #testing
path = kwargs.get('testing_path')
isfolder = kwargs.get('isfolder')
image_list = [path + '/' + f for f in os.listdir(path) if f.endswith('.jpg')] if isfolder else [path]
saver.restore(sess, tf.train.latest_checkpoint(run))
print('Checkpoint restored...')
print('Testing %d images...' % len(image_list))
images = []
predictions = []
time.sleep(0.1)
for i in tqdm.tqdm(range(len(image_list)), unit='images'):
ori_img = misc.imread(image_list[i])
if len(ori_img.shape) < 3:
continue
img = padding(ori_img, 135, 240)
img = np.reshape(img, (1, 135, 240, 3)) / 255.
fd = {x: img, keep_prob: 1}
pred = sess.run(y_pred, feed_dict=fd)
images.append(ori_img)
predictions.append(pred)
time.sleep(0.1)
print('Testing finished!')
for i in range(len(images)):
plt.figure(1)
image = images[i]
sal = np.reshape(predictions[i], (135, 240))
sal = depadding(sal, image.shape[0], image.shape[1])
sal = sal * (sal > np.percentile(sal, 95))
sal = gaussian_filter(sal, sigma=0.09*sal.shape[0])
sal = (sal - np.min(sal)) / (np.max(sal) - np.min(sal))
plt.subplot(211)
plt.imshow(image)
plt.subplot(212)
plt.imshow(sal, cmap='gray')
plt.show()
def begin_train(self):
N_EPOCHS = 30
N_BATCH = 5
N_TRAIN_INS = len(self.train_ending)
best_val_accuracy = 0
best_test_accuracy = 0
test_threshold = 2000.0
prev_percentage = 0.0
speed = 0.0
batch_count = 0.0
for epoch in range(N_EPOCHS):
print "epoch ", epoch,":"
shuffled_index_list = utils.shuffle_index(N_TRAIN_INS)
max_batch = N_TRAIN_INS/N_BATCH
start_time = time.time()
for batch in range(max_batch):
batch_index_list = [shuffled_index_list[i] for i in range(batch * N_BATCH, (batch+1) * N_BATCH)]
train_story = [self.train_story[index] for index in batch_index_list]
train_ending = [self.train_ending[index] for index in batch_index_list]
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
while np.any((np.asarray(batch_index_list) - neg_end_index_list) == 0):
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
neg_end1 = [self.train_ending[index] for index in neg_end_index_list]
train_story_matrix = utils.padding(train_story)
train_ending_matrix = utils.padding(train_ending)
neg_ending1_matrix = utils.padding(neg_end1)
train_story_mask = utils.mask_generator(train_story)
train_ending_mask = utils.mask_generator(train_ending)
neg_ending1_mask = utils.mask_generator(neg_end1)
self.train_func(train_story_matrix, train_story_mask,
train_ending_matrix, train_ending_mask,
neg_ending1_matrix, neg_ending1_mask)
if batch_count != 0 and batch_count % 10 == 0:
speed = N_BATCH * 10.0 / (time.time() - start_time)
start_time = time.time()
percentage = ((batch_count % test_threshold) + 1) / test_threshold * 100
if percentage - prev_percentage >= 1:
utils.progress_bar(percentage, speed)
# peek on val set every 5000 instances(1000 batches)
if batch_count % test_threshold == 0:
if batch_count == 0:
print "initial test"
else:
print" "
print"test on valid set..."
val_result, val_result_list = self.val_set_test()
print "accuracy is: ", val_result*100, "%"
if val_result > best_val_accuracy:
print "new best! test on test set..."
best_val_accuracy = val_result
self.saving_model('val', best_val_accuracy)
pickle.dump(val_result_list, open('./prediction/LSTM_last_1neg_sharewemb_best_val_prediction.pkl','wb'))
test_accuracy, test_result_list = self.test_set_test()
print "test set accuracy: ", test_accuracy * 100, "%"
if test_accuracy > best_test_accuracy:
best_test_accuracy = test_accuracy
print "saving model..."
self.saving_model('test', best_test_accuracy)
pickle.dump(test_result_list, open('./prediction/LSTM_last_1neg_sharewemb_best_test_prediction.pkl','wb'))
batch_count += 1
print "reload best model for testing on test set"
self.reload_model('val')
print "test on test set..."
test_result = self.test_set_test()
print "accuracy is: ", test_result * 100, "%"
def begin_train(self):
N_EPOCHS = 30
N_BATCH = 20
N_TRAIN_INS = len(self.train_ending)
best_val_accuracy = 0
best_test_accuracy = 0
test_threshold = 1000.0
prev_percetage = 0.0
speed = 0.0
batch_count = 0.0
for epoch in range(N_EPOCHS):
print "epoch ", epoch,":"
shuffled_index_list = utils.shuffle_index(N_TRAIN_INS)
max_batch = N_TRAIN_INS/N_BATCH
start_time = time.time()
total_cost = 0.0
total_err_count = 0.0
for batch in range(max_batch):
batch_index_list = [shuffled_index_list[i] for i in range(batch * N_BATCH, (batch+1) * N_BATCH)]
train_story = [self.train_story[index] for index in batch_index_list]
train_ending = [self.train_ending[index] for index in batch_index_list]
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
while np.any((np.asarray(batch_index_list) - neg_end_index_list) == 0):
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
neg_end1 = [self.train_ending[index] for index in neg_end_index_list]
answer = np.random.randint(2, size = N_BATCH)
answer_vec = np.concatenate((answer.reshape(-1,1), (1 - answer).reshape(-1,1)), axis = 1).astype('int64')
train_end1 = []
train_end2 = []
for i in range(N_BATCH):
if answer[i] == 0:
train_end1.append(train_ending[i])
train_end2.append(neg_end1[i])
else:
train_end1.append(neg_end1[i])
train_end2.append(train_ending[i])
train_story_matrix = utils.padding(train_story)
train_end1_matrix = utils.padding(train_end1)
train_end2_matrix = utils.padding(train_end2)
train_story_mask = utils.mask_generator(train_story)
train_end1_mask = utils.mask_generator(train_end1)
train_end2_mask = utils.mask_generator(train_end2)
cost = self.train_func(train_story_matrix, train_story_mask,
train_end1_matrix, train_end1_mask,
train_end2_matrix, train_end2_mask, answer_vec)
prediction = self.prediction(train_story_matrix, train_story_mask,
train_end1_matrix, train_end1_mask,
train_end2_matrix, train_end2_mask)
total_err_count += (prediction - answer).sum()
if batch_count != 0 and batch_count % 10 == 0:
speed = N_BATCH * 10.0 / (time.time() - start_time)
start_time = time.time()
percetage = ((batch_count % test_threshold)+1) / test_threshold * 100
if percetage - prev_percetage >= 1:
utils.progress_bar(percetage, speed)
# peek on val set every 5000 instances(1000 batches)
if batch_count % test_threshold == 0:
if batch_count == 0:
print "initial test"
else:
print" "
print"test on valid set..."
val_result, val_result_list = self.val_set_test()
print "accuracy is: ", val_result*100, "%"
if val_result > best_val_accuracy:
print "new best! test on test set..."
best_val_accuracy = val_result
self.saving_model('val', best_val_accuracy)
pickle.dump(val_result_list, open('./prediction/BLSTM_1neg_class_best_val_prediction.pkl','wb'))
test_accuracy, test_result_list = self.test_set_test()
print "test set accuracy: ", test_accuracy * 100, "%"
if test_accuracy > best_test_accuracy:
best_test_accuracy = test_accuracy
print "saving model..."
self.saving_model('test', best_test_accuracy)
pickle.dump(test_result_list, open('./prediction/BLSTM_1neg_class_best_test_prediction.pkl','wb'))
batch_count += 1
total_cost += cost
accuracy = 1-(total_err_count/(max_batch*N_BATCH))
print ""
print "total cost in this epoch: ", total_cost
print "accuracy in this epoch: ", accuracy * 100, "%"
print "reload best model for testing on test set"
self.reload_model('val')
print "test on test set..."
test_result = self.test_set_test()
print "accuracy is: ", test_result * 100, "%"
def begin_train(self):
N_EPOCHS = 30
N_BATCH = self.batchsize
N_TRAIN_INS = len(self.train_ending)
best_val_accuracy = 0
best_test_accuracy = 0
test_threshold = 5000/N_BATCH
prev_percetage = 0.0
speed = 0.0
batch_count = 0.0
start_batch = 0.0
for epoch in range(N_EPOCHS):
print "epoch ", epoch,":"
shuffled_index_list = utils.shuffle_index(N_TRAIN_INS)
max_batch = N_TRAIN_INS/N_BATCH
start_time = time.time()
total_cost = 0.0
total_err_count = 0.0
for batch in range(max_batch):
batch_index_list = [shuffled_index_list[i] for i in range(batch * N_BATCH, (batch+1) * N_BATCH)]
train_story = [self.train_story[index] for index in batch_index_list]
train_ending = [self.train_ending[index] for index in batch_index_list]
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
while np.any((np.asarray(batch_index_list) - neg_end_index_list) == 0):
neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
neg_end1 = [self.train_ending[index] for index in neg_end_index_list]
answer = np.random.randint(2, size = N_BATCH)
target1 = 1 - answer
target2 = answer
# answer_vec = np.concatenate(((1 - answer).reshape(-1,1), answer.reshape(-1,1)),axis = 1)
end1 = []
end2 = []
for i in range(N_BATCH):
if answer[i] == 0:
end1.append(train_ending[i])
end2.append(neg_end1[i])
else:
end1.append(neg_end1[i])
end2.append(train_ending[i])
train_story_matrix = utils.padding(train_story)
train_end1_matrix = utils.padding(end1)
train_end2_matrix = utils.padding(end2)
train_story_mask = utils.mask_generator(train_story)
train_end1_mask = utils.mask_generator(end1)
train_end2_mask = utils.mask_generator(end2)
cost = self.train_func(train_story_matrix, train_story_mask,
train_end1_matrix, train_end1_mask,
train_end2_matrix, train_end2_mask, target1, target2)
prediction1, prediction2 = self.prediction(train_story_matrix, train_story_mask,
train_end1_matrix, train_end1_mask,
train_end2_matrix, train_end2_mask)
prediction = np.concatenate((np.max(prediction1, axis = 1).reshape(-1,1),
np.max(prediction2, axis = 1).reshape(-1,1)), axis = 1)
total_err_count += abs((np.argmax(prediction, axis = 1) - answer)).sum()
'''master version print'''
percetage = ((batch_count % test_threshold)+1) / test_threshold * 100
if percetage - prev_percetage >= 1:
speed = N_BATCH * (batch_count - start_batch) / (time.time() - start_time)
start_time = time.time()
start_batch = batch_count
utils.progress_bar(percetage, speed)
'''end of print'''
# peek on val set every 5000 instances(1000 batches)
if batch_count % test_threshold == 0:
if batch_count == 0:
print "initial test"
else:
print" "
print"test on valid set..."
val_result, val_result_list = self.val_set_test()
print "accuracy is: ", val_result*100, "%"
if val_result > best_val_accuracy:
print "new best! test on test set..."
best_val_accuracy = val_result
self.saving_model('val', best_val_accuracy)
pickle.dump(val_result_list, open('./prediction/BLSTMLP_'+self.blstmmlp_setting+'_bilinear_'+\
'dropout'+str(self.dropout_rate)+'_batch_'+str(self.batchsize)+'_best_val.pkl','wb'))
test_accuracy, test_result_list = self.test_set_test()
print "test set accuracy: ", test_accuracy * 100, "%"
if test_accuracy > best_test_accuracy:
best_test_accuracy = test_accuracy
print "saving model..."
self.saving_model('test', best_test_accuracy)
pickle.dump(test_result_list, open('./prediction/BLSTMLP_'+self.blstmmlp_setting+'_bilinear_'+\
'dropout'+str(self.dropout_rate)+'_batch_'+str(self.batchsize)+'_best_test.pkl','wb'))
batch_count += 1
total_cost += cost
accuracy = 1.0 - (total_err_count/(max_batch*N_BATCH))
speed = max_batch * N_BATCH / (time.time() - start_time)
print "======================================="
print "epoch summary:"
print "average speed: ", speed, "instances/sec"
print ""
print "total cost in this epoch: ", total_cost
print "accuracy in this epoch: ", accuracy * 100, "%"
print "======================================="
print "reload best model for testing on test set"
self.reload_model('val')
print "test on test set..."
test_result = self.test_set_test()
print "accuracy is: ", test_result * 100, "%"
def encryptPK(self, data, password, sharedKey):
iv = os.urandom(16)
data = padding(data)
key = PBKDF2(sharedKey + password, iv, iterations=10).read(32)
cipher = AES.new(key, AES.MODE_CBC, iv)
return (iv + cipher.encrypt(data)).encode('base64')
def encrypt(self, key, cipherdata):
iv = os.urandom(16)
cipherdata = padding(cipherdata)
key = PBKDF2(key, iv, iterations=10).read(32)
cipher = AES.new(key, AES.MODE_CBC, iv)
return iv + cipher.encrypt(cipherdata)
def begin_train(self):
N_EPOCHS = 30
N_BATCH = self.batchsize
N_TRAIN_INS = len(self.val_answer)
best_val_accuracy = 0
best_test_accuracy = 0
test_threshold = 1000/N_BATCH
prev_percetage = 0.0
speed = 0.0
batch_count = 0.0
start_batch = 0.0
for epoch in range(N_EPOCHS):
print "epoch ", epoch,":"
shuffled_index_list = utils.shuffle_index(N_TRAIN_INS)
max_batch = N_TRAIN_INS/N_BATCH
start_time = time.time()
total_cost = 0.0
total_err_count = 0.0
for batch in range(max_batch):
batch_index_list = [shuffled_index_list[i] for i in range(batch * N_BATCH, (batch+1) * N_BATCH)]
train_story = [[self.val_story[index][i] for index in batch_index_list] for i in range(self.story_nsent)]
train_ending = [self.val_ending1[index] for index in batch_index_list]
# neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
# while np.any((np.asarray(batch_index_list) - neg_end_index_list) == 0):
# neg_end_index_list = np.random.randint(N_TRAIN_INS, size = (N_BATCH,))
# neg_end1 = [self.train_ending[index] for index in neg_end_index_list]
neg_end1 = [self.val_ending2[index] for index in batch_index_list]
# answer = np.random.randint(2, size = N_BATCH)
# target1 = 1 - answer
# target2 = 1 - target1
answer = np.asarray([self.val_answer[index] for index in batch_index_list])
target1 = 1 - answer
target2 = answer
# answer_vec = np.concatenate(((1 - answer).reshape(-1,1), answer.reshape(-1,1)),axis = 1)
end1 = []
end2 = []
# for i in range(N_BATCH):
# if answer[i] == 0:
# end1.append(train_ending[i])
# end2.append(neg_end1[i])
# else:
# end1.append(neg_end1[i])
# end2.append(train_ending[i])
for i in range(N_BATCH):
end1.append(train_ending[i])
end2.append(neg_end1[i])
train_story_matrices = [utils.padding(batch_sent) for batch_sent in train_story]
train_end1_matrix = utils.padding(end1)
train_end2_matrix = utils.padding(end2)
train_story_mask = [utils.mask_generator(batch_sent) for batch_sent in train_story]
train_end1_mask = utils.mask_generator(end1)
train_end2_mask = utils.mask_generator(end2)
cost, prediction1, prediction2 = self.train_func(train_story_matrices[0], train_story_matrices[1], train_story_matrices[2],
train_story_matrices[3], train_end1_matrix, train_end2_matrix,
train_story_mask[0], train_story_mask[1], train_story_mask[2],
train_story_mask[3], train_end1_mask, train_end2_mask, target1, target2)
prediction = np.argmax(np.concatenate((prediction1, prediction2), axis = 1), axis = 1)
predict_answer = np.zeros((N_BATCH, ))
for i in range(N_BATCH):
if prediction[i] == 0:
predict_answer[i] = 1
elif prediction[i] == 1:
predict_answer[i] = 0
elif prediction[i] == 2:
predict_answer[i] = 0
else:
predict_answer[i] = 1
total_err_count += (abs(predict_answer - answer)).sum()
# peek on val set every 5000 instances(1000 batches)
if batch_count % test_threshold == 0:
if batch_count == 0:
print "initial test"
else:
if batch >= 0.0:
print" "
accuracy = 1.0 - (total_err_count/((batch+1)*N_BATCH))
print "training set accuracy: ", accuracy * 100, "%"
print"test on valid set..."
val_result = self.val_set_test()
print "accuracy is: ", val_result*100, "%"
if val_result > best_val_accuracy:
print "new best! test on test set..."
best_val_accuracy = val_result
test_accuracy = self.test_set_test()
print "test set accuracy: ", test_accuracy * 100, "%"
if test_accuracy > best_test_accuracy:
best_test_accuracy = test_accuracy
batch_count += 1.0
'''master version print'''
percetage = ((batch_count % test_threshold )*1.0) / test_threshold * 100
if percetage - prev_percetage >= 1.0:
speed = N_BATCH * (batch_count- start_batch) / (time.time() - start_time)
start_time = time.time()
start_batch = batch_count
utils.progress_bar(percetage, speed)
prev_percetage = percetage
if prev_percetage >= 99:
prev_percetage = 0.0
'''end of print'''
total_cost += cost
print "======================================="
print "epoch summary:"
print "average cost in this epoch: ", total_cost
print "======================================="