00mjk / pjlink

JLink but for python: a link between Mathematica and python with efficient data transfer

‌

PJLink has been superseded by the Wolfram Client for Python

Please use that if you can. If that is insufficient for one reason or another raise an issue there and if they refuse to fix it let me know and I'll try to add it to PJLink when I have the time and energy to.

---

‌

PJLink

PJLink is a package developed to link python and Mathematica. It is currently python 3.3+ only as it turns out to be more challenging to support legacy python2.7 as well in the C-interface than I would currently like, although it will likely be less difficult to do so in the python codebase.

---

‌

Installation

The easiest way to install PJLink is using a paclet server installation:

PacletInstall[
  "PJLink",
  "Site"->
    "http://www.wolframcloud.com/objects/b3m2a1.paclets/PacletServer"
  ]

If you've already installed it you can update using:

PacletUpdate[
  "PJLink",
  "Site"->
    "http://www.wolframcloud.com/objects/b3m2a1.paclets/PacletServer"
  ]

Alternately you can download this repo as a ZIP file and put extract it in $UserBaseDirectory/Applications

---

‌

Usage

Before anything else, we'll load the package:

<<PJLink`

To start you need to call InstallPython to load a python runtime (the first time this is done it may be slow as it compiles the library PJLink uses for communication):

ker=InstallPython[];

After that you can evaluate python code as a string

PyEvaluateString["import numpy as np"];
PyEvaluateString["np.random.rand(5, 5, 5)"]~Short~3

(*Out:*)

{{{0.19340495587982665`,0.04333471882108819`,0.0793674077113492`,0.6746465215828963`,0.9128377509416972`},<<3>>,{0.36109338142334146`,0.44378058193673864`,<<19>>,<<21>>,0.5244287636275622`}},<<3>>,{<<1>>}}

You can also using the PJLink symbolic processing system to evaluate python code directly from Mathematica code. This is the best way to pass data into python:

With[{arr=RandomReal[{-1, 1}, {100, 100,100}]}, PyEvaluate[test=arr,  TimeConstraint->1]]//AbsoluteTiming//First

(*Out:*)

0.102152`

PyEvaluate[test]//Dimensions//AbsoluteTiming

(*Out:*)

{0.020897`,{100,100,100}}

Errors will be returned wrapped in PythonTraceback :

PyEvaluate[nosym]

(*Out:*)

PythonTraceback["Traceback (most recent call last):\n  File \"~/Documents/Python/IDEA/PJLink/PJLink/KernelLink.py\", line 654, in __handleCallPacket\n    self.__callPython()\n  File \"/Users/Mark/Documents/Python/IDEA/PJLink/PJLink/KernelLink.py\", line 1003, in __callPython\n    res = self.__do_call_recursive(pkt)\n File \"~/Documents/Python/IDEA/PJLink/PJLink/KernelLink.py\", line 959, in __do_call_recursive\n    res = eval(arg, self.__EXEC_ENV, self.__EXEC_ENV)\n  File \"<string>\", line 1, in <module>\nNameError: name 'nosym' is not defined\n"]

When done, call ClosePython to clean up the runtime and close the link:

ClosePython[]

You can also use this package to communicate with Mathematica from Python as I detailed here .