.NET Bridge allows Python and R to access .NET libraries, with the .NET library running either locally or on a remote machine. From R or Python one can:
- create .NET objects
- call member functions
- call class functions (i.e. static members)
- access and set properties
- access indexing members
Two packages are provided, in addition to the .NET codebase:
The following data types in arguments are supported:
- .NET objects
- integers (16, 32, 64) bit
- floats (32, 64) bit
- strings
- byte
- enums
- boolean
- arrays of: objects, integers, doubles, boolean, bytes, strings, etc
- vectors (with optional named index)
- matrices (with optional named row and column indices)
Needless to say, the framework supports code written in C#, F#, or other .NET languages. Access to types is made through reflection and does not need to be language aware.
The R or Python packages communicate with the .NET side through simple client / server interactions. Your .NET libraries are loaded by a runner CLRServer.exe that provides a TCP-based API, giving full visibility into your library(ies).
On first use from R or Python, the package will start the .NET bridge server (or alternatively connect to an existing server). If the server is started from within VisualStudio, Xamarin Studio, or other tool, can be run in debug mode, so that you can debug your libraries as they are called from R or Python.
When a method is first called the code looks for all methods in a class that may match based on name and number of arguments and then picks the method from that subset with the closest convertible signature. The argument set need not be a perfect match in terms of types provided that the types can be reasonably converted. For example strings will be converted to enum values if a given signature requires an enum, integers can be converted to floating point, double[] arrays can be applied to double[] or Vector, etc. These signatures are cached so that subsequent calls avoid scanning.
For example if a class has 2 overloaded public methods "F":
public double F (double x, Vector<double> series)public double F (Direction dir, Vector<double> series)
where Direction is enum Direction { Up, Down }. If the object is called from R or python as:
obj.F ('Up', [0.1, 0.2, 3.0, 3.1, 3.2])the second method would be chosen given that 'Up' is convertible to Direction.Up and the numeric array is convertible to Vector<double>.
The .NET bridge server (CLRServer.exe) will not have access to your code unless you indicate a DLL to be loaded. One needs to instruct the server to load a dll or dlls from your environment. There are a number of ways to do this:
- run CLRServer.exe -dll on the command line
- run CLRServer.exe -dll in your favorite IDE (particularly useful for debugging)
- run from within R or Python
From within R one has two options. The first approach is to set an environment variable either within the R session or externally:
## set environment variable
Sys.setenv(rDotNet_DLL="~/Dev/mymodels.dll") OR
Sys.setenv(RDOTNET_DLL="~/Dev/mymodels.dll")
## load package
require(rDotNet)
## create an object and call a method
obj <- .cnew ("NormalDistribution1D", 0.0, 1.0)
obj$F (0.1)The second approch is to explicitly call the rDotNet initialization function:
require(rDotNet)
## initialize
.cinit(dlls="~/Dev/mymodels.dll")
## create an object and call a method
obj <- .cnew ("NormalDistribution1D", 0.0, 1.0)
obj$F (0.1)In python, the CLR is initialized using the CLRApi constructor. .NET object can then be created and interacted with directly:
clr = CLRApi (dll="~/Dev/mymodels.dll")
## create an object and call a method
obj <- clr.new ("NormalDistribution1D", 0.0, 1.0)
obj.F (0.1)Assuming the following (contrived) .NET classes:
namespace com.stg.dummy
{
class Point (double X, double Y);
class Circle
{
Circle (double radius)
{
_radius = radius;
}
public double Radius
{ get { return _radius; } set { _radius = value; } }
public double Area
{ get { return Math.PI * _radius * _radius; } }
...
// function returning list of objects
public List<Point> PointsFor1 (int npoints)
{
var incr = 2.0 * Math.PI / (double)npoints;
var list = new List<Point>[);
for (int i = 0 ; i < npoints ; i++)
{
var theta = (double)i * incr;
var x = _radius * Math.cos(theta);
var y = _radius * Math.sin(theta);
list.Add (new Point(x,y));
}
return list;
}
// function returning array of objects
public Point[] PointsFor2 (int npoints)
{
return PointsFor(npoints).ToArray();
}
}
}The R api uses the $ syntax to reference members much like other R object approaches. Here is how we could call the above from R:
## create circle object
circle <- .cnew("com.stg.dummy.Circle", 10.0)
## get the list of points back
pointlist <- circle$PointsFor1 (100)
## dereference one of the point objects
point <- pointlist[2]
## or do it all in one go
point <- circle$PointsFor1 (100)[3]
## getting a property
circle$Get("Area")
## setting a property
circle$Set("Radius, 20)
The python API provides .NET objects as first class citizens in Python as proxy objects. One can interact with these proxy objects with normal python syntax. Here is how we could call the above from python:
clr = CLRApi.get()
## create circle object
circle = clr.new("com.stg.dummy.Circle", 10.0)
## get the list of points back
pointlist = circle.PointsFor1 (100)
## dereference one of the point objects
point = pointlist[2]
## or do it all in one go
point = circle.PointsFor1 (100)[3]
## getting a property
circle.Area
## setting a property
circle.Radius = 20The R and Python packages were uploaded and approved by CRAN and PyPi respectively. The packages reside at:
- https://cran.r-project.org/package=rDotNet (R package)
- https://pypi.python.org/pypi?name=pydotnet&version=0.9.2&:action=display (Python package)
Please refer to the Installation document.