ME4PyUtils : MATLAB Engine API for Python utils

Aim

This set of python module and matlab functions aim to ease some data type conversion between matlab and numpy python module. The MATLAB(R) Engine API for Python (ME4P) provides a package for Python to call MATLAB as a computational engine. MPE4P provides nativelly constructors to create arrays in Python (matlab.mlarray.X). This type is built on the array python module. Hower it not possible to create directlt and pass numpy.ndarray. This module proposes some tricks that can be used to

• convert such mlarray into numpy.ndarray using frombuffer
• convert numpy.ndarray into matlab.mlarray.X using prealloc and strides
• pass easily sparse matrix as python dict
• avoid to copy data (not always possible with complex)

The conversion to numpy is fast, the bottleneck comes from the data transfert between python and matlab through the MATLAB Engine API for Python when the data become significant.

some tricks and ideas comes from here

Basic Usage

import ME4PyUtils as ME
import numpy as np

# start matlab
eng = matlab.engine.start_matlab()
# conversion to matlab type before using it in matlab
np_a = np.random.uniform(size=(1000))
m_a = ME.np2mlarray(np_a)
eng.std(m_a)
# conversion to numpy type before using it in python
m_b = eng.rand(matlab.int64([1,1000]))
np_b = ME.mlarray2np(m_b)
np.std(np_b)

Note that if copies of the array are needed, it can be done using copy module or np.copy.

Installation

1. install ME4P (see matlab documentation)
2. Just download or clone the project. Eventually unzip the file and add the folder to matlab and python path.

Test and compare speedup with other possibility

python ME4Putils # set speedtest=True

Frequent issues with MATLAB Engine API

MATLAB Engine API for Python dislikes :

• Calling a matlab function with integer parameters instead of float. These parameters are casted into long integer and will mislead matlab with int64

eng.sqrt(1)  # will crash
eng.sqrt(1.) # works
eng.mod(5,2) # works because integer are expected from matlab

• Returning more than one output, without explicitly set nargout

s,id = eng.sort(np2mlarray(np.array([1.1,3.300,2.01])),nargout=2)  # works, id (index) is a double array...
s,id = eng.sort(np2mlarray(np.array([1.1,3.300,2.01])))  # crash

Alternatives

The propose approach use MATLAB Engine API for Python, but other projects have been proposed, and use numpy nativelly.

Transplant (socket communication)

Oct2py (mat-file communication, possible to speed up with RAM drive (tmpfs) by seting temp_dir properly)

mlab