Greetings pytoch community. I am getting an error Expected object of backend CPU but got backend CUDA for argument #2 ‘weight’. As I searched on internet i have added to(device) to my model and my variables var_X_batch aand var_y_batch . But still i am getting error. Any help ?
import numpy as np
X=[]
Y=[]
for i in range(700):
X.append(np.load('X_{}.npy'.format(i)))
Y.append(np.load('Y_{}.npy'.format(i)))
X=np.array(X)
Y=np.array(Y)
#X.shape is (700,448,448,3)
#y.shape is (700,49,5)
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.data
from torch.autograd import Variable
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
def custom_loss(y_pred,y_true):
for t,p in zip(y_true,y_pred):
t=t.view(49,5)
p=p.view(49,5)
idx=torch.unique(t.nonzero()[:,0])
xt=t[idx,0]
xp=p[idx,0]
yt=t[idx,1]
yp=p[idx,1]
eq1=torch.sum((xt-xp)**2 +(yt-yp)**2,dim=0 )
wt=t[idx,2]
wp=p[idx,2]
ht=t[idx,3]
hp=p[idx,3]
eq2=torch.sum((torch.sqrt(torch.abs(wt))-torch.sqrt(torch.abs(wp))**2) + (torch.sqrt(torch.abs(ht))-torch.sqrt(torch.abs(hp))**2))
ct=t[idx,4]
cp=p[idx,4]
eq3=torch.sum((ct-cp)**2)
res=eq1+eq2+10*eq3
print('res',res)
return res
BATCH_SIZE = 5
torch_X = torch.from_numpy(X)
torch_X = torch_X.permute(0, 3, 1, 2)
y=np.reshape(Y,(Y.shape[0],-1))
torch_y = torch.from_numpy(y)
# Pytorch train
train = torch.utils.data.TensorDataset(torch_X,torch_y)
# data loader
train_loader = torch.utils.data.DataLoader(train, batch_size = BATCH_SIZE, shuffle = False)
def fit(model, train_loader):
optimizer = torch.optim.Adam(model.parameters())
EPOCHS = 50
model.train()
for epoch in range(EPOCHS):
correct = 0
for batch_idx, (X_batch, y_batch) in enumerate(train_loader):
var_X_batch = X_batch.to(device)
var_y_batch = y_batch.to(device)
optimizer.zero_grad()
output = model(var_X_batch)
loss = custom_loss(output, var_y_batch)
loss.backward()
optimizer.step()
correct += (output.data == var_y_batch).sum()
if batch_idx % 50 == 0:
print('Epoch : {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}\t Accuracy:{:.3f}%'.format(
epoch, batch_idx*len(X_batch), len(train_loader.dataset), 100.*batch_idx / len(train_loader), loss.data.item(), float(correct*100) / float(BATCH_SIZE*(batch_idx+1))))
class CNN(nn.Module):
def __init__(self):
super(CNN, self).__init__()
self.conv_block = nn.Sequential(
nn.Conv2d(3, 32, kernel_size=3, stride=1, padding=1),
nn.BatchNorm2d(32),
nn.ReLU(inplace=True),
nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2)
)
self.linear_block = nn.Sequential(
nn.Dropout(p=0.5),
nn.Linear(128*112*112, 128),
nn.BatchNorm1d(128),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Linear(128, 64),
nn.BatchNorm1d(64),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Linear(64, 245)
)
def forward(self, x):
x = self.conv_block(x)
print(x.shape)
x = x.view(x.size(0), -1)
x = self.linear_block(x)
return x
cnn = CNN().to(device)
it = iter(train_loader)
X_batch, y_batch = next(it)
print(cnn.forward(X_batch).shape)
fit(cnn,train_loader)