#ifndef KM_H
#define KM_H

#include <stddef.h> // size_t

// forward typedef for callbacks
typedef struct km_rand_src_t_ km_rand_src_t;

// random number source callbacks
typedef struct {
  _Bool (*fill)(km_rand_src_t * const, const size_t, float * const);
  void (*fini)(km_rand_src_t * const);
} km_rand_src_cbs_t;

// random number source
struct km_rand_src_t_ {
  const km_rand_src_cbs_t *cbs;
  void *data;
};

// fill buffer with N random floats
_Bool km_rand_src_fill(km_rand_src_t * const, const size_t, float * const);

// finalize random source
void km_rand_src_fini(km_rand_src_t * const);

// init system random source (uses system rand())
void km_rand_src_system_init(km_rand_src_t *);

// shape of data set
typedef struct {
  size_t num_floats,
         num_ints;
} km_shape_t;

// data set
typedef struct {
  float *floats;
  int *ints;

  km_shape_t shape;

  size_t num_rows,
         capacity;
} km_set_t;

// init data set with shape and initial size
_Bool km_set_init(km_set_t * const, const km_shape_t *, const size_t);
// finalize data set
void km_set_fini(km_set_t * const);

// append rows to data set, growing set if necessary
_Bool km_set_push_rows(km_set_t *, const size_t, const float *, const int *);

// get row from data set
float *km_set_get_row(const km_set_t *, const size_t);

// init a cluster set of size N from a data set by picking random
// cluster centers
_Bool km_clusters_rand_init(
  km_set_t *,
  const size_t,
  const size_t,
  km_rand_src_t *
);

typedef struct {
  void (*on_step)(const km_set_t *, void *);
  void (*on_means)(const km_set_t *, const float *, const size_t, void *);
} km_clusters_solve_cbs_t;

// use k-means to iteratively update the cluster centroids until there
// are no more changes to the centroids
_Bool
km_clusters_solve(
  km_set_t * const,
  const km_set_t * const,
  const km_clusters_solve_cbs_t * const,
  void * const
);

typedef struct {
  const km_set_t * const cluster_set;
  const size_t num_clusters;
  const float mean_distance;
  const size_t num_empty_clusters;
} km_search_row_t;

typedef void (*km_search_row_cb_t)(
  const km_search_row_t * const,
  void *
);

// repeatedly test different cluster sizes and report results
_Bool
km_search(
  const km_set_t * const set,
  const size_t max_clusters,
  const size_t num_tests,
  const km_search_row_cb_t on_row,
  void *cb_data
);

#endif /* KM_H */