Module Algostream_optimization.Objective

What the optimizer maximizes.

Always maximized: a minimizing objective is expressed by negating f. That keeps every comparison in the library a single > with no direction flag to get backwards.

type t = {
  1. name : string;
  2. f : Metrics.t -> float;
    (*

    higher is better

    *)
}
val sharpe : t
val sortino : t
val calmar : t
val ann_return : t
val total_return : t
val min_drawdown : t

-max_drawdown — for a mandate where capital preservation dominates.

val return_over_max_dd : t

Annual return divided by maximum drawdown. Closely related to Calmar but uses arithmetic rather than geometric return, so it is less sensitive to a short sample.

val penalized : base:t -> lambda:float -> t

base − lambda · max_drawdown. The workhorse: a raw Sharpe objective happily selects a configuration that made its money in one lucky stretch and spent the rest underwater.

val require_activity : base:t -> min_trades:int -> min_periods:int -> n_trades:(unit -> int) -> t

Wrap any objective so it scores zero unless the result has at least min_trades trades and min_periods observations. Without this, a configuration that traded twice and got lucky outranks one that traded five hundred times and worked.

val custom : name:string -> f:(Metrics.t -> float) -> t
val score : t -> Metrics.t -> float

Score, mapping a non-finite result to neg_infinity so a degenerate configuration can never win.