Ray: [solved] LZ4 Installation Issues

[Scitator]

Hi,

I am trying to follow the ray tutorials. Nevertheless, I found several bug with save/load functional in ray 0.4 and decide to move to the master brach.

Unfortunately, I again come to bugs, but now with dependencies versions (lz4 package). I tried different versions (1.0.0, 1.1.0, 2.0.0), but all gives the same error:

     36 with ignoring(ImportError):
     37     import lz4
---> 38     compress['lz4'] = lz4.LZ4_compress
     39     decompress['lz4'] = lz4.LZ4_uncompress
     40 

AttributeError: module 'lz4' has no attribute 'LZ4_compress'

So, can you provide package version to reduce number of such issues? And is it correct, that ray version 0.5 would be available soon?

Thanks!

[richardliaw]

Can you post a full stack trace of the error?

Often times when I run into lz4 issues, I can resolve it by running pip install -U dask.

On Tue, Jun 19, 2018 at 11:40 PM Sergey Kolesnikov [e-mail]
wrote:

Hi,

I am trying to follow the ray tutorials
https://github.com/ray-project/tutorial. Nevertheless, I found several
bug with save/load functional in ray 0.4 and decide to move to the master
brach.

Unfortunately, I again come to bugs, but now with dependencies versions
(lz4 package). I tried different versions (1.0.0, 1.1.0, 2.0.0), but all
gives the same error:

 36 with ignoring(ImportError):
 37     import lz4

---> 38 compress['lz4'] = lz4.LZ4_compress
39 decompress['lz4'] = lz4.LZ4_uncompress
40

AttributeError: module 'lz4' has no attribute 'LZ4_compress'

So, can you provide package version to reduce number of such issues? And
is it correct, that ray version 0.5 would be available soon?

Thanks!

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/ray-project/ray/issues/2278, or mute the thread
https://github.com/notifications/unsubscribe-auth/AEUc5VwwM_sHcQiS1uDvxgK8Oh6de4Hzks5t-e5ygaJpZM4UusPE
.

[robertnishihara]

Yeah, that's probably the issue, also see https://github.com/ray-project/ray/issues/2137#issuecomment-391967245.

[Scitator]

Thanks, now all works perfectly.

Just to be sure, I want to implement several RL algos with Ray (I found it really great one with a lot of cool stuff). But with PyTorch neither TensorFlow. So, what is the best way to start with?
Are there any PyTorch examples? (paper says they should be)

[richardliaw]

That's great - there's an example implementation of A3C in PyTorch in the code base here

One thing you could take a look at is trying to implement the standard Policy Gradient in PyTorch. The current tensorflow version is here.

If you have any feedback or questions, feel free to reach out, either via github or on our mailing list!

[robertnishihara]

It's still pretty common to encounter this issue. I wonder if we should include something in RLlib that checks if lz4 is installed and if tensorflow is being used and then raises a better exception.

The issue is where to put the check. We basically want the check to happen iff tf.contrib is being used I think.

Another solution is to remove our usage of tf.contrib.

[silverbacknet]

It looks like this was fixed for good in https://github.com/ray-project/ray/commit/8b6f0d3224055e5e028569e31cfd56316f7ce29e