I am working on a classification algorithm, for which I am training a decision tree in C++, via the mexFunction. I am storing all the nodes of the tree with their respective data points, each node also has some additional information. I have a pointer named "tree" which points/stores all the data. Now I want to get that tree in matlab with all its data, so that I can plot results. Is there a way to copy all the data via the tree pointer and export it to matlab?
depending on many things
- the pointer is the "first" which pointing to a tree or linked list
this case you need to parse the tree or linked list to get all data
- the pointer is pointing to data record ( structure ) this case you need to redefine the data in the struct in desperate data types in MATLAB
You will have to convert our own data structures into MATLAB data structures using "mxCreateNumericArray", "mxGetPr" and friends. Then use "mxArray *plhs[]" of your mex function to return the converted data to MATLAB.
The question is what you want to do with it in matlab. just copying the memory wont be enough since you cant do anything with it in matlab.
If you really need all information in the tree in matlab you need to copy it to structures that matlab can use (one or more of those http://www.mathworks.de/de/help/matlab/create-or-delete-array.html). For that you need to decide how to represent the data in matlab. The pure tree structure for instance could be stored in a single matrix (tree table; each row is a node idx and indizes of left and right child node). How to represent the information in the leaf nodes depends on what you store there.
The other option would be to keep the tree and all of its info in mex and write additional functions that work on the tree and give back only the required results (eg a classification result or something else you want to query, like the number of leaf nodes, etc.). To keep the tree between subsequent calls of the mex function you can either store it in a static variable in the mex file, or pass the point back to matlab cast to a long. Both methods are kind of dirty, but i'm not aware of a different solution.
The question is a little misleading as you cannot import a pointer from C++ to MATLAB but rather you copy all data from your structures to equivalent structures in MATLAB. So you need to write a conversion routine that converts the structure represented by a tree node in C++ to its equivalent in MATLAB. See for example IT++ library for converting basic types from C++ to MATLAB.
For the second option I wrote about I got some sample code wrapping LibConfig into mex on github: https://github.com/pwohlhart/libconfig_mex
Thank you for all the replies, but the issue is just not passing a single structure , the structure has nested structures in it, and its very dense.
You can probably do what you want with Gary Holt's Matwrap (http://lnc.usc.edu/~holt/matwrap/), though I haven't tried it on recent Matlab versions. It provides an easy way for you to bind (SWIG-ish) C++ code, including methods and objects and pointers to them, into the matlab work space. Even if it doesn't quite work for you, the tricks he uses might inspire ways for you to manage C++ pointers to objects in Matlab using MEX.
- Badges
- Science topic