====== NumPy ======
~~NOTOC~~
<block right box round>
**Website**\\
[[http://numpy.scipy.org]]

**NumPy 1.11.2**

**NumExpr 2.6.1**
</block>

===== optimized NUMPY (pip) =====

With preloading flexiblas (wrapper for openblas) you can change the openblas (serial/parallel) during runtime.
<code bash>
LD_PRELOAD=/lib64/libflexiblas.so.3.2 
pip install --no-binary :numpy: numpy
</code>

For updating an exitingn numpy in an virtualenvireonment:

<code bash>
pip install --force-reinstall --ignore-installed --no-binary :numpy: numpy
</code>


===== alternate version of numpy (Intel MKL) =====

This command executes a script loading a special python module providing NumPy 1.7.1, SciPy 0.13.0, Matplotlib 1.1.1rc:

<code bash>
source "/usr/net/bin/module_python"
</code>

The module is only sourced in the current terminal. Start i.e. Spyder (Python IDE) with:

<code bash>
spyder &
</code>

===== alternate version of numpy (1.9.2 optimized) =====

<code bash>
export CFLAGS="-fopenmp -m64 -O3 -Wl,--no-as-needed"
export CXXFLAGS="-fopenmp -m64 -O3 -Wl,--no-as-needed"
export LDFLAGS="-ldl -lm"
export FFLAGS="-fopenmp -m64 -O3"

python setup.py install --prefix=/somewhere
</code>

IF you are on a cluster or on a computer with good hardware, also add -mtune=native as option to CFLAGS,CXXFLAGS and FFLAGS. \\
The module is only sourced in the current terminal. Start i.e. Spyder (Python IDE) with:

<code bash>
spyder &
</code>


===== Tutorials =====
  * [[http://www.loria.fr/~rougier/teaching/numpy.100/index.html|100 NumPy Exercises]]

{{tag>"python"}}