Easily and efficiently share rectangular array (any dimension, any type) or String between Python, Julia, Matlab - Wolfram Language (Mathematica), Java through memory. Each language somehow implements sharing data either for its sub-kernel or the same processes but now you can share between them at a super fast speed. It's backed by a library written in Rust.
To make your life and mine easier, all the function names are the same (snake_case) except Mathematica which does not support underline and is PascalCase and Java which is a convention to be camelCase. All of them have simple logic action_shared_memory[_property], like get_shared_memory_flatten_length in Julia, Python and Matlab which is GetSharedMemoryFlattenLength in Mathematica and all have the same functionality. (Due to Java's strict type syntax, some intermediate methods was introduced)
Let's suppose you have a matrix in julia and you would like to apply Numpy.sin function on it then use Matlab mpower, then in Mathematica use Minors and share that result with Java, how would you do it? Saving in a file? use socket?
If we ignore the Java, just reading and writing a 1000x1000 double matrix to disk on my system took around 8 seconds while using the shared memory it's under 1 second (0.26). It not only helps you to share data, it can share it super fast, far suprior than disk or socket. Here is the steps for a small matrix with this library:
First, we assume, you've downloaded the repository in your C:\download directory.
- Julia
# Setup
include(raw"C:\download\shared_memory\julia\shared_memory.jl")
set_shared_memory_path(raw"C:\download\shared_memory\data")
# Sample Matrix
data = [1 2 3 ; 4 5 6 ; 7 8 9];
# Share
set_shared_memory_data(data) - Python
import numpy
import sys
# Setup
sys.path.append(r'C:\download\shared_memory\python')
from shared_memory import *
set_shared_memory_path(r'C:\download\shared_memory\data')
# Receive
data = get_shared_memory_data()
# Manipulate
new_data = numpy.sin(data);
# Share
set_shared_memory_data(new_data)- Matlab
% Setup
addpath("C:\\download\\shared_memory");
addpath("C:\\download\\shared_memory\\matlab");
set_shared_memory_path("C:\\download\\shared_memory\\data")
% Receive
data = get_shared_memory_data();
% Manipulate
new_data = mpower(data,2);
% Share
set_shared_memory_data(new_data)- Mathematica (Wolfram Language)
(* Setup *)
SharedMemory`libraryPath = "C:\\download\\shared_memory\\shared_memory.dll";
Get["C:\\download\\shared_memory\\mathematica\\shared_memory.wl"];
SetSharedMemoryPath["C:\\download\\shared_memory\\data"];
(* Receive *)
data = GetSharedMemoryData[];
(* Manipulate *)
newData = Minors[data];
(* Share *)
SetSharedMemoryData[newData];- Java
import com.github.ben_izd.shared_memory.SharedMemory;
import java.nio.file.Path;
public class Sample {
public static void main(String[] args) {
// Setup
var libraryPath = Path.of("C:\\download\\shared_memory\\shared_memory.dll");
try(SharedMemory sharedMemory = new SharedMemory(libraryPath)) {
sharedMemory.setSharedMemoryPath("C:\\download\\shared_memory\\data");
// Receive
var data = sharedMemory.getSharedMemoryDouble2D();
}
}
}Assume you have an image inside Mathematica and you want to share it with other interfaces.
First we share our image (stored in sampleImage variable):
(* Setup *)
SharedMemory`libraryPath = "C:\\download\\shared_memory\\shared_memory.dll";
Get["C:\\download\\shared_memory\\mathematica\\shared_memory.wl"]
SetSharedMemoryPath["C:\\download\\shared_memory\\sample_image"]
(* Share *)
SetSharedMemoryData @ NumericArray @ ImageData[sampleImage, "Byte"]% Setup
addpath("C:\\download\\shared_memory");
addpath("C:\\download\\shared_memory\\matlab");
set_shared_memory_path("C:\\download\\shared_memory\\sample_image");
% Receive
data = get_shared_memory_data();
image = permute(data,[3 1 2])# Setup
import sys
sys.path.append(r'C:\download\shared_memory\python')
from shared_memory import *
from PIL import Image
set_shared_memory_path("C:\download\shared_memory\sample_image")
# Receive
data = get_shared_memory_data()
image = Image.fromarray(data, 'RGB')# Setup
include(raw"C:\download\shared_memory\julia\shared_memory.jl")
using .shared_memory
import Images: colorview, RGB
set_shared_memory_path(raw"C:\download\shared_memory\sample_image")
# Receive
data = get_shared_memory_data()
image = colorview(RGB, permutedims(data, [2 3 1]) /255)import com.github.ben_izd.shared_memory.SharedMemory;
import java.nio.file.Path;
import java.awt.image.BufferedImage;
public class Sample {
public static void main(String[] args) throws Throwable {
var libraryPath = Path.of("E:\\projects\\Mathematica\\community\\31. shared_memory\\shared_memory.dll");
try(SharedMemory sharedMemory = new SharedMemory(libraryPath)) {
sharedMemory.setSharedMemoryPath(Path.of("E:\\projects\\Mathematica\\community\\31. shared_memory\\example\\image\\imageData"));
var data = sharedMemory.getSharedMemoryByte3D();
var image = readImage(data);
}
}
public static BufferedImage readImage(byte[][][] data) {
int width = data[0].length;
int height = data.length;
BufferedImage bufferedImage = new BufferedImage(width,height,BufferedImage.TYPE_INT_RGB);
int red,green,blue,rgb;
for (int row = 0; row < height; row++) {
for (int column = 0; column < width; column++) {
red = data[row][column][0] & 0xFF;
green = data[row][column][1] & 0xFF;
blue = data[row][column][2] & 0xFF;
rgb = (red << 16) | (green << 8) | blue;
bufferedImage.setRGB(column,row,rgb);
}
}
return bufferedImage;
}
}matlab folder contains all functions as separate files, just include that folder and the root (in windows folder contains .dll) to path variable with addpath or other methods to be able to use the functions.
Now use set_shared_memory_path function to set your shared memory file, a file used by other programs to access the data you'll share or want to receive.
Advance Tip: If you want to move the shared library (in windows that
shared_memory.dllfile), also move the head fileshared_memory.hwith it.
Before loading the package, first you have to set SharedMemory`libraryPath to the library path (in windows is .dll file). Then, use Get or other functions to load the package.
Just include the shared_memory.jl in your code.
If you moved the shared_memory.jl or shared_memory.dll file, then open shared_memory.jl file and set LIBRARY_PATH to the shared library (in windows is shared_memory.dll file).
You need Numpy installed first.
Add the python folder to your path with sys module and import the file like below:
import sys
sys.path.append(r'C:\download\shared_memory\python')
from shared_memory import *If you moved the shared_memory.py or shared_memory.dll file, then open shared_memory.py file and set LIBRARY_PATH to the shared library (in windows is .dll file).
Add java folder to you class path and give the path of shared library (in windows shared_memory.dll) file to the SharedMemory constructor.
- Only rectangular arrays or single String
- Data stored in shared memory is column-major like Julia and Matlab but Mathematica and python (by default) are row-major, sending and receiving data from Python, Mathematica and Java will apply the necessary transformations.
- Java does not have a complex notion. Because of that, you can't share or receive complex values within Java.
Check all the functions and names in different interfaces in the Wiki section.
The project was built on windows and the dll files is included in the repository, but if you're on other operating systems, you have to build the rust project on your platform which thanks to Rust cargo, is surprisingly easy.
- Install Rust on your computer
- Make sure you have an internet connection and build the Rust project (rust folder) - You can read Building for Release section.
- Go inside target/release, and use the built library.
If you don't have Mathematica, you can comment the mathematica module in lib.rs and also comment wolfram-library-link = ... in cargo.toml file
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent