matSklearn

MATLAB Wrapper for the Python Machine Learning Library scikit-learn (https://github.com/scikit-learn/scikit-learn)

Required Packages

In order to use this library, you will need:

1. MATLAB. I wrote & tested this in version R2016b, it is my assumption this will work in any version of MATLAB that supports calling python libraries. A quick check for this would be to see if the command which py returns a valid path.
2. Python. I wrote & tested this with Python 2.7, you can check (or change) which version your MATLAB is running with the command pyversion. I don't think there's any reason this shouldn't work with Python 3.x, but I just haven't tested it myself. If you need help getting Python & MATLAB working together, this Mathworks page will get you up to speed.
3. Python libraries: numpy, scipy, & scikit-learn. If you don't already have all 3 of these, the easiest thing to do would be to install a product like Anaconda.

Getting Started

1. After you clone the repository, add the matSklearn directory to your MATLAB path using addpath or pathtool.
2. It's possible (but unlikely) that MATLAB found your scikit-learn directory. Try something like this (Windows):

USERPROFILE = regexprep(getenv('USERPROFILE'),'\\','\\\');
PYPACKAGEPATH = strcat(USERPROFILE,'\\AppData\\Local\\Continuum\\Anaconda3\\envs\\py27\\Lib\\site-packages')
if count(py.sys.path,PYPACKAGEPATH) == 0
    insert(py.sys.path,int32(0),PYPACKAGEPATH);
end

Note your PYPACKAGEPATH may be different depending on what version of Python you are using, if you are using a Python IDE other than Anaconda, etc. It's the path to the directory where where pypi or conda put sklearn.

Using matSklearn

For now, only the cluster module has been wrapped. You can see some examples of basic use on a toy dataset in the sklearnClusterUTest m-file.

TODO

I will add a folder for each additional module as they are written. I guess this gives me a hefty #TODO list:

sklearn.cluster

sklearn.calibration
sklearn.cluster.bicluster
sklearn.covariance
sklearn.cross_decomposition
sklearn.datasets
sklearn.decomposition
sklearn.discriminant_analysis
sklearn.dummy
sklearn.ensemble
sklearn.feature_extraction
sklearn.feature_selection
sklearn.gaussian_process
sklearn.isotonic
sklearn.kernel_approximation
sklearn.kernel_ridge
sklearn.linear_model
sklearn.manifold
sklearn.metrics
sklearn.mixture
sklearn.model_selection
sklearn.multiclass
sklearn.multioutput
sklearn.naive_bayes
sklearn.neighbors
sklearn.neural_network
sklearn.pipeline
sklearn.preprocessing
sklearn.random_projection
sklearn.semi_supervised
sklearn.svm
sklearn.tree
sklearn.utilities

I automated the writing of most of the subclass wrappers using a python script (maybe 90% of the writing?). Here's an example:

from sklearn import cluster
import inspect
import re
import networkx as nx


clusterDict = cluster.__dict__
dk = clusterDict.keys()
flavors = filter(lambda k:isinstance(d[k],type),dk)
clusterDict = {}

G = nx.graph.Graph()
G.add_node("sklearn.cluster")
G.add_nodes_from(flavors)

for flavor in flavors:
    G.add_edge("sklearn.cluster",flavor)
    clusterString = "cluster."+flavor+"()"
    print clusterString
    clusterObject = eval(clusterString)
    
    D = dir(clusterObject)
    dictKeys = clusterObject.__dict__.keys()
    D = [d for d in D if d not in dictKeys]
    D = [d for d in D if not re.search('^_+|_+$',d)]
    
    print("\n")
    print(clusterObject.__class__)
    print("\nArgs/Varargs/Kwargs:\n")
    print(inspect.getargspec(clusterObject.__class__.__init__))
    print("\nProperties:\n")
    print(clusterObject.__dict__)
    print("\nMethods:\n")
    print(D)
    print("\n")
    
    clusterObjectDict = {}
    clusterObjectDict['properties'] = clusterObject.__dict__
    clusterObjectDict['methods'] = D
    clusterDict[clusterString] = clusterObjectDict
    
    K = clusterObject.__dict__.keys()
    G.add_nodes_from(K)
    G.add_nodes_from(D)
    for k in K:
        G.add_edge(flavor,k)
    for d in D:
        G.add_edge(flavor,d)

gDegree = G.degree()
nodeList=[k for k in gDegree.iterkeys() if gDegree[k]>2]
#nx.draw_networkx(G,nodelist=nodeList)
nx.draw_networkx(G)

prog = re.compile('(?<=type \')[^\'>]+|(?<=class \')[^\'>]+')
for k1 in clusterDict.iterkeys():
    errMsg = ""
    fptr = open(k1[:-2]+".m",'w')
    fptr.write("classdef "+k1[8:-2]+" < sklearnCluster\n\n")
    fptr.write("% MATLAB wrapper for sklearn."+k1+"\n")
    fptr.write("% Peter Cook 2018\n\n")
    #write class properties
    fptr.write("properties\n")
    for k2 in clusterDict[k1]['properties'].iterkeys():
        #check class of arg/kwarg
        argClass = str(type(clusterDict[k1]['properties'][k2]))
        m = prog.search(argClass)
        argClass = m.group(0)
        argValue = str(clusterDict[k1]['properties'][k2])
        
        if argClass == 'int':
            fptr.write("\t"+k2+" = "+str(argValue)+"\n")
        elif argClass == 'str':
            fptr.write("\t"+k2+" = '"+argValue+"'\n")
        elif argClass == 'float':
            fptr.write("\t"+k2+" = "+str(argValue)+"\n")
        elif argClass == 'bool':
            fptr.write("\t"+k2+" = "+str(argValue).lower()+"\n")
        elif argClass == 'NoneType':
            fptr.write("\t"+k2+" = "+"py.None\n")
        else:
            fptr.write("\t"+k2+" = "+str(argValue)+"\n")
    #end properties
    fptr.write("end\n\n")
    
    #write class methods
    fptr.write("methods\n\n")
    #write class constructor
    fptr.write("function hObj = "+k1[8:-2]+"\n")
    fptr.write("\thObj@sklearnCluster("+k1[8:-2]+")\n")
    fptr.write("end\n\n")
    #write class set methods
    for k2 in clusterDict[k1]['properties'].iterkeys():
        #check class of arg/kwarg
        argClass = str(type(clusterDict[k1]['properties'][k2]))
        m = prog.search(argClass)
        argClass = m.group(0)
        fptr.write("function set."+k2+"(hObj,x)\n")
        if argClass == 'int':
            fptr.write("\tif isnumeric(x) && x>0\n")
            fptr.write("\t\thObj.pyCluster."+k2+" = int16(x);\n")
            fptr.write("\t\thObj."+k2+" = int16(x);\n")

        elif argClass == 'str':
            fptr.write("\tif ischar(x) && any(ismember(x,{}))\n")
            fptr.write("\t\thObj.pyCluster."+k2+" = x;\n")
            fptr.write("\t\thObj."+k2+" = x;\n")
            fptr.write("\t\tCHECK KWARGS\n")
            print("Check kwargs for "+k1+"."+k2)
            errMsg = errMsg+"%Check kwargs for "+k1+"."+k2+"\n"

        elif argClass == 'float':
            fptr.write("\tif isfloat(x) && x>0\n")
            fptr.write("\t\thObj.pyCluster."+k2+" = x;\n")
            fptr.write("\t\thObj."+k2+" = x;\n")

        elif argClass == 'bool':
            fptr.write("\tif islogical(x)\n")
            fptr.write("\t\thObj.pyCluster."+k2+" = x;\n")
            fptr.write("\t\thObj."+k2+" = x;\n")

        elif argClass == 'NoneType':
            fptr.write("\tif isa(x,'py.None')\n")
            fptr.write("\t\tCHECK KWARGS\n")
            print("Check kwargs for "+k1+"."+k2)
            errMsg = errMsg+"%Check kwargs for "+k1+"."+k2+"\n"

        else:
            fptr.write("\tif isa(x,"+argClass+")\n")
            fptr.write("\t\thObj.pyCluster."+k2+" = x;\n")
            fptr.write("\t\thObj."+k2+" = x;\n")
            fptr.write("\t\tCHECK KWARGS\n")
            print("Check kwargs for "+k1+"."+k2)
            errMsg = errMsg+"%Check kwargs for "+k1+"."+k2+"\n"
        
        #end if
        fptr.write("\tend\n")
        #end function
        fptr.write("end"+"\n\n")
    #end methods
    fptr.write("end\n\n")
    #end classdef
    fptr.write(errMsg)
    fptr.write("end\n")
    fptr.close()