Versions for requirements

[josephtey]

I am trying to get the same results as the paper, however, am running into several issues.

I believe this could be due to the version of the required packages; could anyone please list the versions for Python, Theano, Keras, Numpy and Scikit-learn, that work as described in the paper?

Thank you so much, this would be much appreciated.

[wpmarinho]

I used python 2.7 in a Anaconda distribution. theano 0.8.2 and Keras 1.1.0 with a modest Nvidia 930M gpu.
Everything on windows 10

If you are planning to use Windows 10, follow this tutorial on how to set up a training enviroment, it works!:
https://github.com/philferriere/dlwin

I´m studying this algorithm too.

[josephtey]

Hi there,
Sorry for the late response, but still no luck. I think all these packages are fine, but it gives me the error that I do not have tensorflow. I pip installed it, but now i get this error:
Expected int32, got <tensorflow.python.ops.variables.Variable object at 0x111e4e1d0> of type 'Variable' instead.

I'm running this on a Mac, Sierra OS; what could be the issue?

[StarPQ]

I met the same problem as you.
I solve this by switching the backend from tensorflow to theano.
vim $HOME/.keras/keras.json
And change the field backend to "theano", like this
{
"image_data_format": "channels_last",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "theano"
}

You may visit https://keras.io/backend/ for more ditail.

[cloudspurs]

Package Version

---

absl-py 0.7.1
astor 0.7.1
backports.weakref 1.0.post1
enum34 1.1.6
funcsigs 1.0.2
futures 3.2.0
gast 0.2.2
grpcio 1.19.0
h5py 2.9.0
Keras 0.3.3
Keras-Applications 1.0.7
Keras-Preprocessing 1.0.9
Markdown 3.1
mock 2.0.0
numpy 1.10.0
pbr 5.1.3
pip 19.0.3
pkg-resources 0.0.0
protobuf 3.1.0
PyYAML 5.1
scikit-learn 0.20.3
scipy 1.2.1
setuptools 41.0.0
six 1.12.0
sklearn 0.0
tensorboard 1.13.1
tensorflow 0.12.0
tensorflow-estimator 1.13.0
termcolor 1.1.0
Theano 0.8.2
Werkzeug 0.15.2
wheel 0.33.1

[nikhilbalwani]

Can I post a reimplementation of this code which would be consistent with the current versions of these libraries?