Tensor condition slice in libtorch

zhijian_li · August 21, 2020, 12:55am

Hi, I have tensor slicing problem here:
In python:
I can use output[ output < 0.5 ] = 0 to set somevalues to 0
In C++:
I don’t know how to do it.

albanD · August 21, 2020, 1:01am

I think the .index() method is used to mimick the [] behavior.
You might want to check the doc in details though as I’m not sure what this function can do exactly

zhijian_li · August 21, 2020, 1:09am

Thanks for your hint, I’ve checked the docs, there is no condition indexing method found.

albanD · August 21, 2020, 1:11am

Interesting.
Then you’ll have to use masked_fill directly then I guess https://pytorch.org/cppdocs/api/function_namespaceat_1a5118bdabdb9cc642a7a12ceaacfd0531.html?highlight=masked_fill

zhijian_li · August 21, 2020, 1:14am

any example to create a mask?

albanD · August 21, 2020, 1:28am

This is actually creating a boolean Tensor with the mask you want

zhijian_li · August 21, 2020, 1:36am

Thank you, I’ve found the solution, and tested it.
output.index_put_({output < 0.5}, 0)