Intel(R) Python with Intel(R) Optimized Theano

[Christophe_H_Intel2]

For instructions on using Intel's optimized Theano* with your Intel(R) Distribution for Python* installation, please see the article here.

Update:  Beginning with 2017 update 2, the Intel Theano module is included in the Intel Python distribution.

Update:  As of 2017 update 3, it is no longer necessary to use a .theanorc file. The optimal defaults are determined at runtime.

[raymond_s_2]

The theanorc file recommended in the article requires an intel compiler. If you're using a gcc compiler try this,

nano ~/.theanorc

[cuda]
root = /usr/local/cuda

[global]
device = cpu
floatX = float32
mode = FAST_RUN
openmp = True
openmp_elemwise_minsize = 10

[gcc]
cxxflags = -O3 -ffast-math -ftree-loop-distribution -funroll-loops -ftracer

[blas]
ldflags = -lmkl_rt

[Priyadhars_K_Intel]

Hi,

Do we have Keras on Theano ?

if someone is using Theano on a GPGPU , how to optimize their workload

Regards,

Priya 

[RL5]

raymond s. wrote:

The theanorc file recommended in the article requires an intel compiler. If you're using a gcc compiler try this,

Thank you for the addition. In my case I don't see a difference between using the default cxx flags or using the ones suggested in this post.

[Christophe_H_Intel2]

Hi RL,

Beginning with Intel Python 2017 Update 3, the .theanorc file is no longer required.  The recommended defaults, including environment variables for Xeon Phi, are set at runtime based on the cpu and installed compiler.   Additionally, the steps decribed in the article above for installing Intel Theano are no longer necessary, as the module is bundled with Intel Python.

Thanks,

Chris