DOCUMENTATION.md

Scikit-learn API Reference

To learn more about the scikit-learn API click here.

c-scikit-learn implementation

The folder structure of c-scikit-learn is the following:

c-skl
│   skl_helper.h
│   skl_helper.c
│
└───linear_model
│   │
│   └───linear_regression
│       │   skl_linear_regression.h
│       │   skl_linear_regression.c
│       │
└───────svm
        │
        └────svr
             │   skl_svr.h
             │   skl_svc
    ...

Each folder in c-skl contains a distinct model from scikit-learn.

array datatype

c-scikit-learn makes use of a double** wrapper, which is used by every model in c-scikit-learn.

typedef struct array{
  int r;
  int c;
  double** x;
} array;

Model structure

A model has the following structure:

#ifndef SKL_<MODEL_NAME>
#define SKL_<MODEL_NAME>

#define NO_IMPORT_ARRAY
#include "../../skl_helper.h"

typedef struct skl_<model_name> skl_<model_name>;

// skl_<model_name> model struct
struct skl_<model_name> {
	PyObject* self; // reference to a scikit-learn object
	struct {
	   int parameter_1;
	   int parameter_2;
	   .
	   .
	   .
	   char* parameter_n;
	} parameters;
	struct {
	   int attribute_1;
	   int attribute_2;
	   .
	   .
	   .
	   array* attribute_3;
	} attributes;
	// methods associated to <model_name>
 	void (*fit)(skl_<model_name>* m, array* x, array* y);
 	void (*get_params)(skl_<model_name>* m);
 	array* (*predict)(skl_<model_name>* m, array* x);
 	float (*score)(skl_<model_name>* m, array* x, array* y);
 	void (*set_params)(skl_<model_name>* m);
 	void (*purge)(skl_<model_name>* m);
};

// fetches a <model_name> struct with default parameters
skl_<model_name>* skl_get_<model_name>();

#endif /* SKL_<MODEL_NAME> */

Other than that, the functionality is pretty much the same as in scikit-learn.

For an example, check out a toy linear regression example.