C11/C++ library for least-squares polynomial regression. Aims to have a pleasant, portable API with plenty of fluff.

1. maxOrder can be increased, but bounded by the stack
2. Buffers are not properly freed when errors arise
3. Lacks strerror
4. Elides multiple error states
5. See open issue

Fits an N-order polynomial to a set of input points (see point set abstraction).

Fits an N-order polynomial to a set of input points.

Returns a human-readable string describing a polyfit status code.

Writes into "buf" a string representing a polynomial.

Computes the value of a polynomial expression for a given x

While optimal for many use cases, the const double *xs, const double *ys signature may cause unecessary allocations and/or memcpy calls. Consider the case where you have an array of vector structs:

typedef struct {
    double x;
    double y;
} v2;

v2 arr[N_POINTS];

This array is fundamentally incompatible with the aforementioned signature. Instead of mangling this data into the format demanded by the signature, we can instead use the pf_points_t abstraction:

struct pf_points_by_v2s {
    pf_points_t super;
    const v2 *v2s;
};

static double pf_points_by_v2s_x(const pf_points_t *self, pf_len_t idx) {
    return ((struct pf_points_by_v2s *) self)->v2s[idx].x;
}

static double pf_points_by_v2s_y(const pf_points_t *self, pf_len_t idx) {
    return ((struct pf_points_by_v2s *) self)->v2s[idx].y;
}

//

struct pf_points_by_v2s points_impl;
points_impl.super.count = N_POINTS;
points_impl.super.x = pf_points_by_v2s_x;
points_impl.super.y = pf_points_by_v2s_y;
points_impl.v2s = arr;

pf_points_t *points = &points_impl.super;
// pass "points" to pf_fit

This involves 0 heap allocations, and is a reasonable tradeoff given that each point is only read approximately once by the internal algorithm. This also means that the x/y accessor may be fairly expensive while incurring no additional penalty.