pip install error

[fpreimesberger]

With pip install --user google-nucleus it gives the error

Could not find a version that satisfies the requirement google-nucleus (from versions: )
No matching distribution found for google-nucleus

Is there a fix for this? Machine is 18.04 Ubuntu

[ThomasColthurst]

Interesting! Are you able to pip install other packages? Also, are you by chance running a Python 3 version of pip? (pip --version will tell you.)

Right now, the installed version of Nucleus is Python 3 compatible, but the pip package is Python 2 only. We hope to fix that soon.

[ThomasColthurst]

Closing for now. Please re-open with version info if problem persists.

[ajlogo]

I would like to be a beta tester of a python/pip 3 install of Nucleus.
The Nucleus pip install instructions should include the condition of python 2 only.
I found this page after doing a search, and others may just stop at the pip 3 error.

I was able to install a python 2.7 environment inside of Anaconda 3 (Anaconda 3 is an anaconda version and does not imply python 3 environments only).
Installing the nucleus with pip worked.
Got the notebook to work locally with CPU.
The GPU version runs but repeats/recreates the following about 200 times:
Created TensorFlow device
full message:
2019-02-06 17:29:30.017084: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 6999 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080, pci bus id: 0000:0a:00.0, compute capability: 6.1)

This may just be related to the number of batches.
The result is that the GPU time is about the same as the CPU time.

[TimSmole]

@ThomasColthurst is there any new information on when we can expect Nucleus to be available via pip for Python3 as well?

Thanks!

[ThomasColthurst]

A Python 3 pip for Nucleus is probably a few weeks away.

[TimSmole]

@ThomasColthurst Perfect! Thanks for quick response. Also one more question - will Nucleus be compatible with Tensorflow 2.0 in the near future?

[ajlogo]

Thomas, Thank you very much

[ThomasColthurst]

@TimSmole -- yes. We have an upcoming Nucleus release that removes our dependencies on TensorFlow's Python code, and this will make it easy to use Nucleus with whatever version of Tensorflow you want.

(Nucleus's C++ code will still depend on a specific version of TensorFlow, and that specific version will get upgraded to TensorFlow 2.0 once 2.0 is out of alpha. But that shouldn't effect most Nucleus users.)

[TimSmole]

Great news! Thanks again @ThomasColthurst! I'm looking forward to this release.

[ThomasColthurst]

The long awaited Python 3-compatible pip package for Nucleus is now available.