Up
# Module Make (Key : Piecewise_linear_intf.Float_like) (Value : Piecewise_linear_intf.Float_like) : S with type
key = Key.t with type
value = Value.t

### Parameters

### Signature

type
key
=
Key.t

type
value
=
Value.t

`create`

enforces that x (key) values are non-decreasing.

It also enforces certain finiteness conditions: the x and y values must be finite (non-nan, and non-infinite), and differences of consecutive x values and consecutive y values must be finite.

`get t x`

evaluates the piecewise linear function `t`

at `x`

.

It is possible to get discontinuous functions by using repeated x-values in the
knots. In that case, the function is evaluated in such a way that it is
right-continuous. For example, if `t`

has knots
`[(0.,0.5); (1.,1.5); (1.,10.); (2.,11.)]`

, then `get t 1.`

returns `10.`

,
`get t 0.999`

returns `1.499`

, and `get t 1.001`

returns `10.001`

.

val
precache : ?density:float -> t -> unit

`precache t`

computes and stores a lookup table in `t`

that speeds up subsequent
calls to `get t`

. Any call to `get`

needs to find the knots that define the
interval in which the key lies. This is done by bisection. Ordinarily the
bisection starts on the whole domain of the piecewise linear function. Precaching
builds a lookup table based on an equispaced division of the domain. This allows
`get`

to quickly determine a (potentially very) small initial interval on which to
start the bisection.

This works best for knots that are reasonably evenly distributed.

`density`

is the ratio of the size of the lookup table to the size of the knot
array.

Calling `precache`

multiple times is safe. If the existing lookup density is the
same or higher density than the requested density, the lookup table will not be
recomputed.

Returns the `t`

such that `get t key`

= sum (`get t_i key`

) * `weight_i`

. This will
fail if given an empty list as an argument, if any weights are not finite, or if
any of the input `t`

s has a discontinuity.

The domain of each `t`

does not have to be the same. The domain of the `t`

that is
returned will be the connected union of the domains.

There are cases in `S_invertible`

in which all `t`

s could be valid and invertible,
but the linear combination is not invertible. I.e. if one `t`

is downward sloping,
the other `t`

is upward sloping, and the linear combination is sometimes upward and
sometimes downward sloping.