Loss function with targets of type list of list of tensors

fastai
1

Hello, everyone. I am new to fastAI, sorry for the silly question.
I am trying to use lr_find with my network model. My loss function receives, as targets, a list of list of tensor. Unfortunately I am geting this error message:

File “trainFastai.py”, line 38, in
main()
File “trainFastai.py”, line 32, in main
fastai.train.lr_find(learner, start_lr=0.00001, end_lr=1.0, num_it=len(trainDataset)*10)
File “/usr/local/lib/python3.6/dist-packages/fastai/train.py”, line 41, in lr_find
learn.fit(epochs, start_lr, callbacks=[cb], wd=wd)
File “/usr/local/lib/python3.6/dist-packages/fastai/basic_train.py”, line 200, in fit
fit(epochs, self, metrics=self.metrics, callbacks=self.callbacks+callbacks)
File “/usr/local/lib/python3.6/dist-packages/fastai/basic_train.py”, line 101, in fit
loss = loss_batch(learn.model, xb, yb, learn.loss_func, learn.opt, cb_handler)
File “/usr/local/lib/python3.6/dist-packages/fastai/basic_train.py”, line 30, in loss_batch
loss = loss_func(out, *yb)
File “/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py”, line 493, in call
result = self.forward(*input, **kwargs)
TypeError: forward() takes 3 positional arguments but 12 were given

I think that Learner is unpacking the target lists. How should I work around this problem?

2

You should define your loss function to get that list unpacked. If it’s always the same size, you can name the arguments, if not, you can use a star:

def my_loss_func(out, *targets):

and targets will be a list.

3

Thanks for your reply @sgugger!
Using the star worked. Can you explain why I need this star?
Thanks again!

4

That’s just python syntax: *args means any number of arguments.