def epoch_loop():
train_parameters = config.select(
'model_type, learning_rate, which_text') + {
'param': 'train_loss',
'time': time.ctime()
}
val_parameters = config.select('model_type, learning_rate, which_text') + {
'param': 'val_loss',
'time': time.ctime()
}
# register parameters for hyperboard
train_loss_agent = agent.register(train_parameters.getdict(),
'loss',
overwrite=True)
val_loss_agent = agent.register(val_parameters.getdict(),
'loss',
overwrite=True)
gvar = GlobalVariable()
trainDataset = ArchDataset(config.copy(train=True), gvar)
validDataset = ArchDataset(config.copy(train=False), gvar)
checkpoints = config.select('model_type, learning_rate, which_text') + {
'train_loss': [],
'valid_loss': []
}
checkpoints.update(start_time=time.asctime())
for epoch in range(config.max_epoch):
model.train()
train_loss = train(epoch, trainDataset)
agent.append(train_loss_agent, epoch, train_loss)
checkpoints.train_loss.append(train_loss)
model.eval()
val_loss = validation(epoch, validDataset)
agent.append(val_loss_agent, epoch, val_loss)
checkpoints.valid_loss.append(val_loss)
if (epoch + 1) % config.savefreq == 0:
print('saving model')
#checkpoints.update(model = model)
checkpoints.update(end_time=time.asctime())
torch.save(model, config.checkpoint)
#coding=utf-8
from torch.autograd import Variable
from configure import GlobalVariable
from PIL import Image
import numpy as np
import json
import urllib
gvar = GlobalVariable()
gvar.resnet = gvar.resnet.cuda()
data_pair = [
json.loads(line.strip()) for line in open('static/dataset/data_pair.json')
]
query_num = 10
def handle_sentence(query_sen):
dists = np.square(gvar.sentences - query_sen).sum(1)
index = dists.argsort()
dists = dists[index]
topk = [data_pair[index[i]] for i in range(query_num)]
print(topk)
images = ['static/dataset/' + topk[i]['imgpath'] for i in range(query_num)]
for i in range(query_num):
for key in topk[i]:
topk[i][key] = str(topk[i][key])
return topk, images
'''
# get images
import numpy as np
get_ipython().magic('matplotlib inline')
plt.plot(title_chars_lengths)
#plt.show()
plt.plot(title_words_lengths)
plt.show()
# In[17]:
from configure import GlobalVariable
gvar = GlobalVariable()
print gvar.
# In[101]:
import random
import numpy as np
def getitem(index):
title = data_pair[index]['keywords']
sentence = gvar.sentence_transform(title)
if sentence.shape[0] == 0:
sentence = np.zeros((1, 300))
return sentence.mean(0)
# coding: utf-8
# In[1]:
from configure import GlobalVariable
from linecache_light import LineCache
gvar = GlobalVariable()
data_pair = LineCache('static/dataset/data_pair.json')
print gvar.arch_feats.shape
print gvar.sentence_ids.shape, gvar.sentence_ids
print data_pair.num_lines
# In[2]:
import numpy as np
arch_feats = np.load('images.npy')
print arch_feats.shape
from torchvision import transforms
img_transform = transforms.Compose([
transforms.Scale(100),
transforms.CenterCrop(100),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224,
0.225]),
])
# In[3]: