Notes on Macbook Air -- Using Intel HD 5000

[Dexdev08]

I've managed to install tf-coriander in OSX mavericks using the wheel file. Thanks a lot!

Don't know where to put this - should i put this in the wiki? The least I can do to document my experience for anyone who will come to this package. Looking forward to keras v2.0 compatibility.

Keras ver 1.1.1 TF v0.11.0rc0

I realized the documentation in keras.io currently is already for the new version, as following examples there may not be updated.

example: documentation: keras.utils.to_categorical(y, num_classes) model.fit - uses epochs as a parameter

keras v1.1.1 keras.utils.np_utils.to_categorical(y, nb_classes) model.fit - uses nb_epoch as a parameter

[hughperkins]

Yes, wiki sounds good :-)

[Dexdev08]

OK! Just saved it. Will close this.

[hughperkins]

You mentioned in the other thread that it could be interesting to have keras 1.1.1 doc? To what extent is https://faroit.github.io/keras-docs/1.1.1/ approximately what you are looking for?

[Dexdev08]

Yes, this is what I was looking for. Thanks! I will add it to the wiki page.

[hughperkins]

cool :-)

[Dexdev08]

Also a question - I'm going through all examples now - some examples, I encounter errors, but some seem to run just fine. Where do you think I should log these? (Also do you think it's worth to just upgrade my OS version to Sierra since Apple just stopped supporting this version last December?)

[hughperkins]

Can you paste the full output, including the command you are running, into a gist, and paste into a comment in this thread, one gist per failing example?

[Dexdev08]

1_Introduction/helloworld.py

https://gist.github.com/Dexdev08/576ff5a701878f18c42ec6ee49c85233

[Dexdev08]

3_NeuralNetworks/bidirectional_rnn.py

https://gist.github.com/Dexdev08/338dd6aca84a042e23e47ff39ebc7554

[hughperkins]

For helloworld.py, helloworld.py is not supported, since it contains strings. => normal, expected result

[hughperkins]

For bidirectional_rnn.py, can you define the envireonmetn variable:

export COCL_OFFSETS_32BIT=1

... and retry please? More information: https://github.com/hughperkins/coriander/blob/master/doc/options.md#cocl_offsets_32bit-for-beignet

[Dexdev08]

I think it has something to do with easycl?

https://gist.github.com/Dexdev08/5149c9097afe0d7df441c4d0b6cd834b

[hughperkins]

Background: easycl is a wrapper around OpenCL. It is handling storing the failed kernel for us, and detecging the error, but the undelrying error is probably because buffer offsets are sent to the kernels as 64-bit integers by default, which beignet isnt a big fan of. See https://github.com/hughperkins/tf-coriander/issues/55#issuecomment-310842616 , above

[Dexdev08]

3_Neural_Networks/convolutional https://gist.github.com/Dexdev08/fe27566859e8c088161f3ff6cf9d88c6

[Dexdev08]

I think only autoencoders work for my case - if it's because of the OS then i'll probably just upgrade in the next few days. If it is a hardware limitation - at least good to know that I should acquire new hardware.

[hughperkins]

Please set the environment variable as above, and then retry

On 24 June 2017 15:57:01 BST, Dexdev08 notifications@github.com wrote:

I think only autoencoders work for my case - if it's because of the OS then i'll probably just upgrade in the next few days. If it is a hardware limitation - at least good to know that I should acquire new hardware.

-- You are receiving this because you commented. Reply to this email directly or view it on GitHub: https://github.com/hughperkins/tf-coriander/issues/55#issuecomment-310843292

-- Sent from my Android device with K-9 Mail. Please excuse my brevity.

[Dexdev08]

3_neural_networks_convolutional_network_COCL_error https://gist.github.com/Dexdev08/d84fb4c4daa853e994493b8273ba3fcc

[Dexdev08]

This is the example from the keras documentation which worked https://gist.github.com/Dexdev08/0f01bf364e47d860443690c120410268

[hughperkins]

Please set the env var, and retry bidirectional rnn

On 24 June 2017 16:05:32 BST, Dexdev08 notifications@github.com wrote:

This is the example from the keras documentation which worked https://gist.github.com/Dexdev08/0f01bf364e47d860443690c120410268

-- You are receiving this because you commented. Reply to this email directly or view it on GitHub: https://github.com/hughperkins/tf-coriander/issues/55#issuecomment-310843807

-- Sent from my Android device with K-9 Mail. Please excuse my brevity.

[Dexdev08]

Hi Hugh, this is the gist with the env var. https://gist.github.com/Dexdev08/5149c9097afe0d7df441c4d0b6cd834b

[Dexdev08]

Hi Hugh!

I've finally upgraded my laptop to sierra and was able to run the tensorflow examples! I would recommend to Macbook Air 2013 users to upgrade to Sierra.

Still got errors on the convolutional network... the rest seems to work.

Does this mean i can happily run more stuff on this?

Regards, Dexter

[Dexdev08]

Just an update Hugh - I've tried a cpu version vs gpu version, the1.5GB memory and 384mb GPU memory of the macbook air is severely limiting for my experiments.

[hughperkins]

Right :) I'll close this then?

[Dexdev08]

Yes. Sad though i wanted to be an opencl proponent. - but im not even knowledgable enouhh to help you out.

On Mon, 21 Aug 2017 at 16:00, Hugh Perkins notifications@github.com wrote:

Right :) I'll close this then?

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/hughperkins/tf-coriander/issues/55#issuecomment-323674033, or mute the thread https://github.com/notifications/unsubscribe-auth/AEc4Dv3oVQNeIxjCKHV8NKUsBgdJTpG3ks5saTkwgaJpZM4OEU1y .