Microsoft Azure Data Science Virtual Machine with Theano and Keras




Before getting started, make sure you have the following:

The data science virtual machine (DSVM) on Azure, based on Windows Server 2012, contains popular tools for data science modelling and development activities such as Microsoft R Server Developer Edition, Anaconda Python, Jupyter notebooks for Python and R, Visual Studio Community Edition with Python and R Tools, Power BI desktop, SQL Server Developer edition, and many other data science and ML tools.

The DSVM is the perfect tool for academic educators and researchers and will  jump-start modelling and development for your data science project.

This deep learning toolkit provides GPU versions of mxnet and CNTK for use on Azure GPU N-series instances.

These Nvidia GPUs use discrete device assignment, resulting in performance that is close to bare-metal, and are well-suited to deep learning problems that require large training sets and expensive computational training efforts. The deep learning toolkit also provides a set of sample deep learning solutions that use the GPU, including image recognition on the CIFAR-10 database and a character recognition sample on the MNIST database.

Deploying this toolkit requires access to Azure GPU NC-class instances for GPU based workloads.

Setting up CUDA and GCC

CUDA, go the NVIDIA’s website and download the CUDA 7.5 toolkit. Select the right version for you computer. 
The GCC compiler, there are many GCC compliers available TDM-gcc has a 64 bit version

To make sure that everything is working at this point, run the the following command on the command line (cmd.exe) .

where gcc

where cl

where nvcc

where cudafe

where cudafe++

If if finds the path for everything you are good to go.

Getting Theano :

In DSVM, we have anaconda python 2.7 installed. Anaconda Python installs all dependencies of Theano.

Step 1: Open Windows Anaconda prompt and execute the following in command line –

conda install mingw libpython

Step 2: Installing Theano – Once the dependencies are installed, you can download and install Theano. To get the latest bleeding edge version go to Theano on GitHub and download the latest zip. Then unzip it somewhere.

Step 3: Configuring Theano – Once you have downloaded and unzipped Theano, cd to the folder and run

python develop

This step will add the Theano directory to your PYTHON_PATH environment variable.

Step 4: Getting Keras – Keras runs on Theano by default, so we just get it via pip

pip install keras

After installing the python libraries you need to tell Theano to use the GPU instead of the CPU.  A lot of older posts would have you set this in the system environment, but it is possible to make a config file in your home directory named “.theanorc.txt” instead.  This also makes it easy to switch out config files.  Inside the file put the following:

device = gpu
floatX = float32

compiler_bindir=C:Program Files (x86)Microsoft Visual Studio 12.0VCbin

Lastly, set up the Keras config file ~/.keras/keras.json.  If you haven’t started Keras yet, the folder and file won’t be there but you can create it. Inside the config put the following.

 "image_dim_ordering": "tf",
 "epsilon": 1e-07,
 "floatx": "float32", 
 "backend": "theano"

Testing Theano with GPU

To  check if your installation of Theano is using your GPU use the following code 

from theano import function, config, shared, sandbox

import theano.tensor as T

import numpy

import time

vlen = 10 * 30 * 768 # 10 x #cores x # threads per core

iters = 1000

rng = numpy.random.RandomState(22)

x = shared(numpy.asarray(rng.rand(vlen), config.floatX))

f = function([], T.exp(x))


t0 = time.time()

for i in range(iters):

r = f()

t1 = time.time()

print("Looping %d times took %f seconds" % (iters, t1 - t0))

print("Result is %s" % (r,))

if numpy.any([isinstance(x.op, T.Elemwise) for x in f.maker.fgraph.toposort()]):

print('Used the cpu')


print('Used the gpu')

Testing Keras with GPU

This code will make sure that everything is working and train a model on some random data. The first time might take a little longer because it the software needs to do some compiling.

from keras.models import Sequential

from keras.layers import Dense, Activation

# for a single-input model with 2 classes (binary):

model = Sequential()

model.add(Dense(1, input_dim=784, activation='sigmoid'))




# generate dummy data

import numpy as np

data = np.random.random((1000, 784))

labels = np.random.randint(2, size=(1000, 1))

# train the model, iterating on the data in batches

# of 32 samples, labels, nb_epoch=10, batch_size=32)

If everything works you will see



This tutorial is an improved version which allows you to make Theano and Keras work with Python 3.5 on Windows 10 if you have access to a local PC with a Nvidia GPU.

This sample shows how to import Theano and Keras into Azure ML and use them in Execute Python Script module. 

Keras Documentation 

Theano Documentation 

Using Tensorflow on Azure see