When trying to accumulate gradients in the manner shown below a runtime error occurs.
print("Training...\n")
model.train()
for epoch in tqdm(range(args.num_epochs)):
cost = 0
hidden_state = None
optimizer.zero_grad()
for index in range(train_features.shape[0]):
print("🔴 Index: {}".format(index))
# y_hat, hidden_state = model(G, features, edge_weights, hidden_state)
y_hat = model(train_features[index], edge_weights)
cost = cost + torch.mean((y_hat-train_targets[index])**2)
cost = cost / (index+1)
cost.backward()
optimizer.step()
The runtime error raised is as follows
File "/usr/local/lib/python3.8/dist-packages/seastar-0.1-py3.8.egg/seastar/executor.py", line 269, in <listcomp>
ret = tuple([self.ts.tensor_map[grad.id] if grad != None else None for grad in arg_grads] + [None for grad in ret_grads])
KeyError: 'V9'
When trying to accumulate gradients in the manner shown below a runtime error occurs.
The runtime error raised is as follows