How to use Mathematica functions in Python programs?

Question

I'd like to know how can I call Mathematica functions from Python.

I appreciate a example, for example, using the Mathematica function Prime.

I had search about MathLink but how to use it in Python is a little obscure to me.

I tried to use a Mathematica-Python library called pyml but I hadn't no sucess, maybe because this lib looks very old (in tutorial says Mathematica 2 or 3).

So, someone knows a good way to write python programs who uses Mathematica functions and can give me an example?

Old Edit:

Maybe this edit can help someone who wants to use mathlinks directly.

To another solution, please see the answer accepted.

Using the Wolfram/Mathematica/8.0/SystemFiles/Links/Python I could had sucess in compiling the module changing some things in setup.py.

My architechture is x86-64.

1-Change the mathematicaversion to 8.0.

2-Changing the lib name ML32i3 to ML64i3.

3-Copying the file Wolfram/Mathematica/8.0/SystemFiles/Libraries/Linux-x86-64/libML64i3.so to the path pointed in setup.py library_dirs = ["/usr/local/Wolfram/Mathematica/" + mathematicaversion + "/SystemFiles/Links/MathLink/DeveloperKit/Linux/CompilerAdditions"].

5-Compiling the source with sudo python setup.py build.

6-Installing the lib with sudo python setup.py install

4-Editing the file /etc/ld.so.conf and putting the line include /usr/local/lib.

5-Creating a directory in /usr/local/lib/python2.6/dist-packages/mathlink with the lib libML64i3.so.

6-Running sudo ldconfig

I had tested the scripts guifrontend.py with python guifrontend.py -linkname "math -mathlink" -linkmode launch and textfrontend.py with python textfrontend.py -linkname "math -mathlink" -linkmode launch and worked fine.

Looks like I almost. But the script

>>> from mathlink import *
>>> import exceptions,sys, re, os
>>> from types import ListType
>>> mathematicaversion = "8.0"
>>> os.environ["PATH"] = "/usr/local/Wolfram/Mathematica/" + mathematicaversion + ":/usr/local/bin:/usr/bin:/bin"
>>> e = env()
>>> sys.argv=['textfrontend.py', '-linkname', 'math -mathlink', '-linkmode', 'launch']
>>> kernel = e.openargv(sys.argv)
>>> kernel.connect()
>>> kernel.ready()
0
>>> kernel.putfunction("Prime",1)
>>> kernel.putinteger(10)
>>> kernel.flush()
>>> kernel.ready()
0
>>> kernel.nextpacket()
8
>>> packetdescriptiondictionary[3]
'ReturnPacket'
>>> kernel.getinteger()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
mathlink.error: MLGet out of sequence.

breaks in the last command and I don't know why. How can I fix this?

Answer 1

This solution can work with several programming languages. Check this GitHub repository of mine.

See this link.

I have found a solution. Works fine to me.

Steps:

1-Create a script named runMath with the content:

#!/usr/bin/env wolframscript
# for certain older versions of Mathematica replace 'wolframscript' by
# 'MathematicaScript -script' in the shebang line

value=ToExpression[$ScriptCommandLine[[2]]];

(*The next line prints the script name.*)
(*Print[$ScriptCommandLine[[1]]];*)

Print[value];

2-I gave execution privilege to the file.

sudo chmod +x runMath

3-Moved the file to the execution path

sudo mv runMath /usr/local/bin/

4-Created a new script called run with the content:

#!/usr/bin/python
from subprocess import *
from sys import *

command='/usr/local/bin/runMath'
parameter=argv[1]

call([command,parameter])

5-Moved to the execution path

sudo mv run /usr/local/bin

6-Finally, tested it:

$run Prime[100]
541

$run 'Sum[2x-1,{x,1,k}]'
k^2

$run Integrate[Log[x],x]
-x + x*Log[x]

$run 'Zeta[2]'
Pi^2/6

You can use with ou without '. The ' are needed to command with spaces.

$run 'f[n_] := f[n] = f[n - 1] + f[n - 2]; f[1] = f[2] = 1; Table[f[n],{n,5}]'
{1, 1, 2, 3, 5}

Happy!

Answer 2

You can call Mathematica function in Python using the Python MathLink module (the source you found in .../SystemFiles/Links/Python), though you'll need to edit a couple of setup files to get it up and running ([email protected] should be able to help you out there).

To use Prime from Python you would run something like:

kernel.ready()

0

kernel.putfunction("Prime",1)

kernel.putinteger(10)

kernel.flush()

kernel.ready()

1

kernel.nextpacket()

3

packetdescriptiondictionary[3]

'ReturnPacket'

kernel.getinteger()

29

Answer 3

Better late than never, but the error in the python script appears to be due to the format of the current packet. Using your routine (with k = kernel):

>>> k.putfunction("Prime",1)
>>> k.putinteger(10)
>>> k.flush()
>>> k.nextpacket()
8

8 is a RETURNPKT, but what is the kernel returning?

>>> k.getnext()
34
>>> tokendictionary[34]
'MLTKSTR'

The kernel wants to send you a string, let's grab that instead

>>> k.getstring()
'In[1]:= '

I have noticed this behavior with the Python MathLink, namely that the kernel pretty adamantly wants to send us the first input prompt. Fair enough. Continuing:

>>> k. nextpacket()
3
>>> tokendictionary[k.getnext()]
'MLTKINT'
>>> k.getinteger()
29

Finally. The lesson here is that the python MathLink bindings need the end-user to pay very close attention to the packets being sent and their type. It is also very unforgiving; once you have thrown an error, no other kernel requests will be honored until k.clearerror() is called.

Answer 4

Try Jupyter and SAGE notebook extensions for Mathematica:

http://www.sagemath.org/doc/reference/interfaces/sage/interfaces/mathematica.html

https://github.com/bjedwards/IPython-mathematicamagic-extension

Answer 5

As yet another solution, you can use a new package called PJLink which implements JLink for python.

Once you have the package installed you start a KernelLink object and can use that to either evaluate strings or use the MathematicaBlock helper to write Mathematica-esque code:

>>> import sys, os
>>> path_to_PJLink = os.path.expanduser("~/Documents/Python/IDEA/PJLink")
>>> sys.path.insert(0, path_to_PJLink)
>>> from PJLink import create_kernel_link
>>> kernel = create_kernel_link()
>>> kernel.evaluateString("$Version")
'11.3.0 for Mac OS X x86 (64-bit) (March 7, 2018)'
>>> with MathematicaBlock():
...   print(kernel.evaluate(RandomReal([-1, 1], [5, 5, 5])))
... 
[[[ 0.82333428 -0.22120173  0.28668275  0.04229312  0.92312944]
  [-0.43474731  0.64218193 -0.54186876 -0.6428228  -0.21671452]
  [-0.05066144 -0.29888543 -0.45723433  0.46926194 -0.72166916]
  [ 0.60358171  0.67254163  0.22736352  0.42711063  0.702733  ]
  [ 0.89118955 -0.76092357  0.65784946  0.42381519 -0.18184603]]

 [[-0.05566714  0.32324674  0.16537235  0.29394423 -0.3826456 ]
  [-0.30799836  0.61072882  0.87721626 -0.40292037 -0.99221593]
  [ 0.79848099  0.86318783 -0.09653227 -0.30479808  0.81070852]
  [ 0.15083946  0.71684487 -0.54012157 -0.70812674  0.16847963]
  [-0.54408339 -0.17084765  0.25754549  0.62201424 -0.41126613]]

 [[ 0.24181817  0.65082232 -0.9448459  -0.6740693  -0.70014564]
  [ 0.04181131 -0.45243563  0.02266605  0.67429721  0.53853541]
  [ 0.79842224  0.38250834 -0.42321046  0.62054957 -0.78189105]
  [ 0.91608362  0.91750963 -0.03043691  0.35382865 -0.42971801]
  [ 0.05856562  0.38297739  0.21494046 -0.77326953 -0.06090548]]

 [[-0.38581506  0.40810805 -0.06970923  0.58814432  0.970209  ]
  [-0.63803971 -0.26755306 -0.55287498  0.51696225  0.2504991 ]
  [ 0.57078906 -0.40804371 -0.26543291 -0.05022896 -0.34215896]
  [-0.27644055 -0.06931716  0.17401444 -0.5905933  -0.74513478]
  [ 0.16364204 -0.12781597 -0.77921589  0.90835378 -0.86846471]]

 [[-0.00391625  0.64797141 -0.32733043 -0.85628204 -0.39096691]
  [ 0.99579966 -0.08433519 -0.94709096  0.42123235 -0.35505469]
  [ 0.62157133  0.85142865  0.63162374 -0.43469092 -0.86983996]
  [ 0.98171147  0.0102183  -0.45457086 -0.40369601  0.45952086]
  [ 0.38977525 -0.23563561 -0.66688691  0.51853403  0.23597912]]]

that last bit there returned as a numpy.ndarray since I had NumPy installed, but it uses a different memoryview-esque thing if there's no NumPy to use

Answer 6

I haven't done this before, so can't share any useful experiences, but you can try using JLink from Jython to connect to Mathematica. I am fairly sure that the amount of glue code necessary will be quite small.

A quick hack to make your attempt for PyhtonLink work would be to add the folder with the mentioned shared lib to LD_LIBRARY_PATH (Linux). You can do so by modifying some global scripts, or you can start your process in a shell script, having first redefined LD_LIBRARY_PATH there.

Answer 7

In May 16, 2019, Wolfram Client Library for Python is announced, so nowadays the problem can be solved easily with:

from wolframclient.evaluation import WolframLanguageSession
from wolframclient.language import wl

session = WolframLanguageSession()

session.evaluate(wl.Prime(50))
#229