Letianwu / ZMCintegral

An easy way to use multi-GPUs to calculate multi-dimensional integration
https://arxiv.org/pdf/1902.07916v2.pdf
Apache License 2.0
19 stars 8 forks source link

TensorFlow version #15

Open jonas-eschle opened 5 years ago

jonas-eschle commented 5 years ago

Hi, I've read your paper with great interest, but I could not find the TensorFlow version of ZMCintegral. Could you may point me to the right repository?

JCatesPH commented 5 years ago

@Letianwu I have the same question as above!

jonas-eschle commented 5 years ago

I've asked the author ZHANG Junjie directly, I am posting here his reply:

The tensorflow version is here: https://github.com/Letianwu/ZMCintegral/tree/version-2.0/version2.1. We are still working on this package such that it could be more flexible and efficient in the future. The current version, if is going to be used directly, require some understanding of the code. Later, we will make this package more user friendly with our new algrothm involved.

JCatesPH commented 5 years ago

Thank you! @mayou36

I wonder if I can avoid some of the headache coming from the cuda compilation issues now.

jonas-eschle commented 5 years ago

@JCatesPH there are a lot of reasons in my experience, anyway I think it has to do with TensorFlow, right? There are tons of ways to install TensorFlow (recommending Anaconda), but it depends what you specific do. Search around, I am sure you'll find something. If not, you may open a new issue and post it there

JCatesPH commented 5 years ago

@mayou36 I meant that I have been working on an integration with ZMCintegral using the numba version for a little while. There has been one thing after another whenever solving a different problem, such as the usage of arrays of variable size in the numba cuda environment. The "typing" and "constant inference" issues in numba cuda has plagued me the entire time. I also had to write new versions of the Bessel functions, a Heaviside function, and other helper functions just to get numba.cuda to understand any of it.

Anyway, I got the script to run on the hpc yesterday finally, but it didn't seem to be completing the integration sampling.

jonas-eschle commented 5 years ago

I see, but then you may can post this as a separate issue, I think this is the best thing

mahmoudasmar commented 5 years ago

Can we have an example of how to use the tensor-flow version of the code, as in the Cuda version. Thanks

JCatesPH commented 5 years ago

@mahmoudasmar There is an example on this page: https://github.com/Letianwu/ZMCintegral/tree/version-2.0/version2.1

It seems even more straightforward than the numba version.

jonas-eschle commented 4 years ago

Is there any updates on the TensorFlow version of this package?

jonas-eschle commented 3 years ago

Hi all, just asking back, are there any news on the TensorFlow version?