return (img_as_tensor, single_image_label)
def __len__(self):
return len(self.data.index)
img_transform = transforms.Compose([transforms.ToTensor()])
dataset = CustomDatasetFromCSV(
'/home/17mcmi06/harsha/project/draw_tel_unbalanced/UHTelPCC.csv',
32,
32,
transform=img_transform)
train_loader = DataLoader(dataset, batch_size=batch_size, shuffle=True)
model = DrawModel(T, A, B, z_size, N, dec_size, enc_size)
optimizer = optim.Adam(model.parameters(),
lr=learning_rate,
betas=(beta1, 0.999))
if USE_CUDA:
model.cuda()
def generate_image(count):
x = model.generate(batch_size)
return x
def save_example_image():
from config import *
from utility import Variable, save_image, xrecons_grid
import torch.nn.utils
import matplotlib.pyplot as plt
torch.set_default_tensor_type('torch.FloatTensor')
train_loader = torch.utils.data.DataLoader(datasets.MNIST(
'../data',
train=True,
download=True,
transform=transforms.Compose([transforms.ToTensor()])),
batch_size=batch_size,
shuffle=False)
model = DrawModel(T, A, B, z_size, N, dec_size, enc_size, use_att)
optimizer = optim.Adam(model.parameters(),
lr=learning_rate,
betas=(beta1, 0.999))
if USE_CUDA:
model.cuda()
def train():
avg_loss = 0
count = 0
for epoch in range(epoch_num):
for data, _ in train_loader:
bs = data.size()[0]
data = Variable(data).view(bs, -1)
from draw_model import DrawModel
from config import *
#from torchvision.utils import save_image
from utility import *
import torch.nn.utils
import numpy as np
import pandas as pd
torch.set_default_tensor_type('torch.FloatTensor')
model = DrawModel(T,A,B,z_size,N,dec_size,enc_size)
if USE_CUDA:
model.cuda()
state_dict = torch.load('save/weights_15600.tar')
model.load_state_dict(state_dict)
def save_imag(x,i):
for t in range(T):
img = xrecons_grid(x[t],B,A)
plt.matshow(img, cmap=plt.cm.gray)
imgname = 'gen_image/i_%d_%s_%d.png' % (i,'gen', t) # you can merge using imagemagick, i.e. convert -delay 10 -loop 0 *.png mnist.gif
plt.savefig(imgname)
print(imgname)
def generate(epoch):
x,_ = model.generate(batch_size)
return x
#save_imag(x,epoch)
import torch.utils
from draw_model import DrawModel
from config import *
from utility import Variable,save_image,xrecons_grid
import torch.nn.utils
import matplotlib.pyplot as plt
torch.set_default_tensor_type('torch.FloatTensor')
train_loader = torch.utils.data.DataLoader(
datasets.MNIST('data/', train=True, download=True,
transform=transforms.Compose([
transforms.ToTensor()])),
batch_size=batch_size, shuffle=False)
model = DrawModel(T,A,B,z_size,N,dec_size,enc_size)
optimizer = optim.Adam(model.parameters(),lr=learning_rate,betas=(beta1,0.999))
if USE_CUDA:
model.cuda()
def train():
avg_loss = 0
count = 0
for epoch in range(epoch_num):
for data,_ in train_loader:
bs = data.size()[0]
data = Variable(data).view(bs, -1)
optimizer.zero_grad()
loss = model.loss(data)
avg_loss += loss.cpu().data.numpy()