Hello together,
I’m trying to train a GNN using the MNIST dataset which I previously turn into a slic segmentation using the Compose class:
import torch
import torchvision.datasets as dset
import torchvision.transforms as transforms
import torchvision.transforms as T
from torch_geometric.transforms import ToSLIC
trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))])
trans = T.Compose([T.ToTensor(), ToSLIC(n_segments=75)])
train_set = dset.MNIST(root='/data', train=True, transform=trans, download=True)
test_set = dset.MNIST(root='/data', train=False, transform=trans, download=True)
batch_size = 100
Afterwards I try to train my model however I have issues that I get receive the error that I’m passing an object and not a tensor. My model training looks somewhat like this:
from torch.nn import Linear
import torch.nn.functional as F
from torch_geometric.nn import GCNConv, TopKPooling, global_mean_pool
from torch_geometric.nn import global_mean_pool as gap, global_max_pool as gmp
embedding_size = 64
class GCN(torch.nn.Module):
def __init__(self):
super(GCN, self).__init__()
torch.manual_seed(42)
self.init_conv = GCNConv(num_features, 64)
self.conv1 = GCNConv(num_features, 64)
self.conv2 = GCNConv(num_features, 64)
self.conv3 = GCNConv(num_features, 64)
self.out = Linear(embedding_size*2, 10)
def forward(self, x, edge_index, batch_index):
hidden = self.initial_conv(x, edge_index)
hidden = F.tanh(hidden)
hidden = self.conv1(hidden, edge_index)
hidden = F.tanh(hidden)
hidden = self.conv2(hidden, edge_index)
hidden = F.tanh(hidden)
hidden = self.conv3(hidden, edge_index)
hidden = F.tanh(hidden)
hidden = torch.cat([gmp(hidden, batch_index),
gap(hidden, batch_index)])
out = self.out(hidden)
return out, hidden
model = GCN()
loss = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(),lr=0.0001)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = model.to(device)
data_size = len(data)
NUM_PER_BATCH = 32
train_loader = torch.utils.data.DataLoader(data[:int(data_size*0.8)],
batch_size = NUM_PER_BATCH, shuffle=True)
test_loader = torch.utils.data.DataLoader(data[int(data_size*0.8):],
batch_size = NUM_PER_BATCH, shuffle=True)
def train(train_loader):
for batch in train_loader:
batch.to(device)
print(batch)
optimizer.zero_grad()
pred, embedding = model(batch.x.float(), batch.pos.float(), batch.batch)
loss = torch.sqrt(loss(pred, batch.y))
loss.backward()
optimizer.step()
return loss, embedding
loss = []
for epochs in range(10):
loss, h = train(train_loader)
losses.append(loss)
if epoch % 2 == 0:
print("Loss: {}".format(loss))
What I also noticed is that ToSlic does not store the edge_index which I assume has to be precalculated (and might cause the error). For guidance I would be very grateful.