About an errror during training process

[Hongzunhei]

Hi, I encountered the following error while trying to port the nashmtl method into my own multi-task project. Could you please help me understand how to resolve it?
Traceback (most recent call last):
File "/home/data1/train_3_tasks_loss3.py", line 1376, in
main()
File "/home/data1/train_3_tasks_loss3.py", line 628, in main
train_metrics = train_one_epoch(
File "/home/data1/train_3_tasks_loss3.py", line 912, in train_one_epoch
) = _forward()
File "/home/data1/train_3_tasks_loss3.py", line 833, in _forward
loss, extra_outputs = weight_method.backward(
File "/home/data1/lib/methods/weight_methods.py", line 1326, in backward
return self.method.backward(losses, **kwargs)
File "/home/data1/lib/methods/weight_methods.py", line 68, in backward
loss, extra_outputs = self.get_weighted_loss(
File "/home/data1/lib/methods/weight_methods.py", line 286, in get_weighted_loss
alpha = self.solve_optimization(GTG.cpu().detach().numpy())
File "/home/data1/lib/methods/weight_methods.py", line 190, in solve_optimization
self.alpha_param.value = alpha_t
File "/home/anaconda/envs/voxelmorph/lib/python3.8/site-packages/cvxpy/expressions/leaf.py", line 436, in value
self.save_value(self._validate_value(val))
File "/home/anaconda/envs/voxelmorph/lib/python3.8/site-packages/cvxpy/expressions/leaf.py", line 510, in _validate_value
raise ValueError(
ValueError: Variable value must be nonnegative.

It seems there is an issue with setting a nonnegative value for a variable. Any suggestions on how to fix this?

Thank you!

[AvivNavon]

Hi @Hongzunhei, could you please provide more information and some additional logs during runtime to try and pinpoint the issue?