====== Deep Learning (DL) ======


===== DL and DataLAB community =====

The [[https://colab.tuwien.ac.at/display/DLDC/|DLDC platform]] is designed to foster the exchange of scientists from different fields in DL. There you will find community meeting announcements and the slides of past meetings. In addition, it is a space to host
moderated special interest groups in the field of DL.

===== DL modules on VSC-3  =====

==== Tensorflow ====
Typically on some GPU node,
<code>
[someone@n372-003 ~]$  module purge 
[someone@n372-003 ~]$  module load gcc/5.3 mpioff/1.0 bazel/0.12.0 cuda/9.0.176 intel-mkl/2018 hdf5/1.8.18-SERIAL python/3.6 numpy/1.14.3 h5py/2.7.0 scipy/0.18.0 tensorflow/1.8.0.ttest 
</code>

==== Keras ====
Typically on some GPU node,
<code>
[someone@n372-003 ~]$  module purge 
[someone@n372-003 ~]$  module load gcc/5.3 mpioff/1.0 bazel/0.12.0 cuda/9.0.176 intel-mkl/2018 hdf5/1.8.18-SERIAL python/3.6 numpy/1.14.3 h5py/2.7.0 scipy/0.18.0 tensorflow/1.8.0.ttest keras/2.1.6.ttest
</code>

==== Pytorch ====
Typically on some GPU node,
<code>
[someone@n372-003 ~]$  module purge 
[someone@n372-003 ~]$  module load gcc/5.3 cuda/9.1.85 intel-mkl/2018 cmake/3.9.6 python/3.6 numpy/1.14.3 pytorch/1.1.0b
</code>

==== Theano ====
Typically on some GPU node,
<code>
[someone@n372-003 ~]$  module purge 
[someone@n372-003 ~]$  module load gcc/5.3 cuda/9.1.85 intel-mkl/11 cmake/3.8.2 python/2.7 numpy/1.12.0 libgpuarray/0.7.6 scipy/0.17.1 Theano/0.9.0
[someone@n372-003 ~]$  export MKL_THREADING_LAYER=GNU
[someone@n372-003 ~]$  THEANO_FLAGS=device=cuda,floatX=float32 python whatevermyscript.py
</code>

Many users nowadays prefer conda-based installs of DL software. For example on some GPU node,
<code>
[someone@n372-003 ~]$  module purge 
[someone@n372-003 ~]$  module load anaconda3/5.3.0
[someone@n372-003 ~]$  module show anaconda3/5.3.0 
[someone@n372-003 ~]$  source /opt/sw/x86_64/glibc-2.17/ivybridge-ep/anaconda3/5.3.0/etc/profile.d/conda.sh
[someone@n372-003 ~]$  conda activate base
(base) [someone@n372-003 ~]$  conda list   ( just to see what packages are currently around )
(base) [someone@n372-003 ~]$  conda search tensorflow
(base) [someone@n372-003 ~]$  conda create -n mytf "tensorflow 1.14.0 gpu_py37h4491b45_0"
                                 --> y   ( say yes to the suggested list of packages to install )
(base) [someone@n372-003 ~]$  conda deactivate    ( to get out of the 'base' environment again )
[someone@n372-003 ~]$  conda info --envs
                          --> mytf      /home/lv123456/someone/.conda/envs/mytf
[someone@n372-003 ~]$  conda activate mytf
(mytf) [someone@n372-003 ~]$  python
Python 3.7.blablabla
>>> import tensorflow as tf    ( lots of FutureWarning... but probably not harmful !? )
>>> hello = tf.constant('Hello, TensorFlow!')
>>> sess = tf.Session()
>>> print(sess.run(hello))
>>> <CTRL>D
(mytf) [someone@n372-003 ~]$  conda deactivate
</code>