theano tutorial is broken about cleverhans

[goodfeli]

@mahnerak can you fix this?

[mahnerak]

Using TensorFlow backend.

You need to somehow specify which backend to use for Keras. You have to either edit your keras.json config file or run every time like this
KERAS_BACKEND=theano python mnist_tutorial_th.py

[goodfeli]

Because the tutorial works only with the theano backend, you should use an if starement and change the backend if necessary, just as the tensorflow tutorial does

[goodfeli]

Or at least give a useful error message

[mahnerak]

Actually, the TensorFlow version of tutorial changes only image dimension ordering, not a backend for Keras.
If you run mnist_tutorial_tf.py with Keras configured theano as default backend, you'll get

Using Theano backend.
Using gpu device 0: GeForce GTX 980 (CNMeM is disabled, cuDNN 5005)
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcublas.so.7.5 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcufft.so.7.5 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:125] successfully opened CUDA library libcurand.so.7.5 locally
INFO: '~/.keras/keras.json' sets 'image_dim_ordering' to 'th', temporarily setting to 'tf'
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
F tensorflow/stream_executor/cuda/cuda_driver.cc:334] current context was not created by the StreamExecutor cuda_driver API: 0x254a160; a CUDA runtime call was likely performed without using a StreamExecutor context
[1] 30067 abort (core dumped) KERAS_BACKEND=theano python mnist_tutorial_tf.py

[mahnerak]

Which one do you prefer? To assert that backend is correct, or change it manually?
If we decide to change backend during runtime, we have to do something like this before importing Keras
os.environ['KERAS_BACKEND']='tensorflow'
Don't know if it's a good idea.

[goodfeli]

OK, I see the issue. I'll take care of this, actually.

…

On Fri, Jan 20, 2017 at 8:52 AM, Karen Hambardzumyan < ***@***.***> wrote: Which one do you prefer? To assert that backend is correct, or change it manually. If we decide to change backend during runtime, we have to do something like this before importing Keras os.environ['KERAS_BACKEND']='tensorflow' — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#60 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAXrGr4mFOKwZJA-TZPKlZ1O6Rt9N3ulks5rUOZCgaJpZM4Low3E> .