neural_network.mli

module Layer : sig
  type t

  val of_size : int -> t
  val to_size : t -> int
end

module Parameter : sig
  module Node_from : sig
    type t =
      | Bias
      | Index of int
    [@@deriving sexp,compare,hash]
  end

  module Node_to : sig
    type t =
      | Index of int
    [@@deriving sexp,compare,hash]
  end

  type t =
    { layer_from_index : int
    ; node_from : Node_from.t
    ; node_to : Node_to.t
    } [@@deriving sexp,compare,hash]

  include Core.Hashable.S with type t := t
end

type t =
  { layers : Layer.t list
  ; parameters : Parameter.t list
  ; index_of_parameter : Parameter.t -> int option
  ; parameterized_output : float list -> Autodiff.Float.t list
  }

val create_exn
  :  ?activation:Autodiff.Float.Univar.t
  -> (int * int) list list
  -> t

val create_full_exn
  :  ?activation:Autodiff.Float.Univar.t
  -> int list
  -> t

module Method : sig
  type t =
    | Newton
    | Gradient_descent_with_step_size of float
  [@@ deriving sexp]

  val arg : t Core.Command.Arg_type.t
end

val train_parameters
  :  ?cost_of_output_and_answer:Autodiff.Float.t
  -> ?regularization:float
  -> ?epsilon_init:float
  -> ?robust:bool
  -> ?method_:Method.t
  -> ?iterations:int
  -> t
  -> inputs_and_answers:(float list * float list) list
  -> float list * Newton.Status.t

val output
  :  t
  -> trained_parameters:float list
  -> (float list -> float list) Core.Staged.t