Got internal error in __sub: no metatable error when I train the sample dataset

[naobit]

I followed Getting Started instruction and installed pix2pix.
Then, downloaded facades dataset by bash ./datasets/download_dataset.sh facades
and run DATA_ROOT=./datasets/facades name=facades_generation which_direction=BtoA th train.lua

but I got the following error.

~/torch-cl/install/bin/luajit: ./models.lua:69: internal error in __sub: no metatable
stack traceback:
	[C]: in function '__sub'
	./models.lua:69: in function 'defineG_unet'
	train.lua:110: in function 'defineG'
	train.lua:146: in main chunk
	[C]: in function 'dofile'
	...i/torch-cl/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
	[C]: at 0x010ec81ce0

How can I fix this?

• run on Mac OSX
• python2.7

[limadm]

I'm investigating this issue and it seems to have its roots in a torch-cl file. I registered an issue there.

[Quasimondo]

I am getting the same error. Is there any temporary workaround?

[phillipi]

Thanks for the pointer @limadm. Maybe it would work to rewrite defineG_unet without the __sub syntactic sugar, instead using the __call__ operator? You can see examples of how to do this here. Another workaround would be to rewrite defineG_unet using nn rather than nngraph, but that will be a bit cumbersome.

[Quasimondo]

So the workaround I found is to reactivate regular torch by reinstalling it. I didn't realize that if you install cl-torch it becomes the default torch. I'm pretty sure there must be a simpler way to switch between the distros, at least this comment suggests it: hughperkins/distro-cl#1 - but this line:

source ~/torch/install/bin/torch-activate
luajit -l cunn -e 'nn.testcuda()'

gave me some errors (but I might have messed up my torch installation at some earlier point)

[phillipi]

Great!

[limadm]

You can use cl-torch if you don't use this metamethod, like I did in this fork.
(it is what you said @phillipi)

[hj0921]

@naobit what method have you used to solve this problem? Thank you!