Optimise.Generic Functor

This document is auto-generated for Owl’s APIs. #60 entries have been extracted. timestamp: 2018-04-16 13:12:54

Github: {Signature} {Implementation}

Utils module

val sample_num : t -> int

Return the total number of samples in passed in ndarray.

source code

val draw_samples : t -> t -> int -> t * t

draw_samples x y draws samples from both x (observations) and y (labels). The samples will be drew along axis 0, so x and y must agree along axis 0.

source code

val get_chunk : t -> t -> int -> int -> t * t

get_chunk x y i c gets a continuous chunk of c samples from position i from x (observations) and y (labels).

source code

Learning_Rate module

type typ =
  | Adagrad of float
  | Const of float
  | Decay of float * float
  | Exp_decay of float * float
  | RMSprop of float * float
  | Adam of float * float * float
  | Schedule of float array

types of learning rate

val run : typ -> int -> t -> t array -> t

Execute the computations defined in module typ.

val default : typ -> typ

Create module typ with default values.

val update_ch : typ -> t -> t array -> t array

Update the cache of gradients.

source code

val to_string : typ -> string

Convert the module typ to its string representation.

Batch module

type typ = Full | Mini of int | Sample of int | Stochastic

Types of batches.

val run : typ -> t -> t -> int -> t * t

Execute the computations defined in module typ.

val batches : typ -> t -> int

Return the total number of batches given a batch typ.

val to_string : typ -> string

Convert the module typ to its string representation.

Loss module

type typ =
  | Hinge
  | L1norm
  | L2norm
  | Quadratic
  | Cross_entropy
  | Custom of (t -> t -> t)

Types of loss functions.

val run : typ -> t -> t -> t

Execute the computations defined in module typ.

val to_string : typ -> string

Convert the module typ to its string representation.

Gradient module

type typ = GD | CG | CD | NonlinearCG | DaiYuanCG | NewtonCG | Newton

Types of gradient function.

val run : typ -> (t -> t) -> t -> t -> t -> t -> t

Execute the computations defined in module typ.

val to_string : typ -> string

Convert the module typ to its string representation.

Momentum module

type typ = Standard of float | Nesterov of float | None

Types of momentum functions.

val run : typ -> t -> t -> t

Execute the computations defined in module typ.

val default : typ -> typ

Create module typ with default values.

val to_string : typ -> string

Convert the module typ to its string representation.

Regularisation module

type typ =
  | L1norm of float
  | L2norm of float
  | Elastic_net of float * float
  | None

Types of regularisation functions.

val run : typ -> t -> t

Execute the computations defined in module typ.

val to_string : typ -> string

Convert the module typ to its string representation.

Clipping module

type typ = L2norm of float | Value of float * float | None

Types of clipping functions.

val run : typ -> t -> t

Execute the computations defined in module typ.

val default : typ -> typ

Create module typ with default values.

val to_string : typ -> string

Convert the module typ to its string representation.

Stopping module

type typ = Const of float | Early of int * int | None

Types of stopping functions.

val run : typ -> float -> bool

Execute the computations defined in module typ.

val default : typ -> typ

Create module typ with default values.

val to_string : typ -> string

Convert the module typ to its string representation.

Checkpoint module

type state = {
  mutable current_batch : int;
  mutable batches_per_epoch : int;
  mutable epochs : float;
  mutable batches : int;
  mutable loss : t array;
  mutable start_at : float;
  mutable stop : bool;
  mutable gs : t array array;
  mutable ps : t array array;
  mutable us : t array array;
  mutable ch : t array array array;
  }

Type definition of checkpoint

type typ =
  | Batch of int
  | Epoch of float
  | Custom of (state -> unit)
  | None

Batch type.

val init_state : int -> float -> state

init_state batches_per_epoch epochs initialises a state by specifying the number of batches per epoch and the number of epochs in total.

source code

val default_checkpoint_fun : (string -> 'a) -> 'a

This function is used for saving intermediate files during optimisation.

source code

val print_state_info : state -> unit

Print out the detail information of current state.

source code

val print_summary : state -> unit

Print out the summary of current state.

source code

val run : typ -> (string -> unit) -> int -> t -> state -> unit

Execute the computations defined in module typ.

val to_string : typ -> string

Convert the module typ to its string representation.

Params module

type typ = {
  mutable epochs : float;
  mutable batch : Batch.typ;
  mutable gradient : Gradient.typ;
  mutable loss : Loss.typ;
  mutable learning_rate : Learning_Rate.typ;
  mutable regularisation : Regularisation.typ;
  mutable momentum : Momentum.typ;
  mutable clipping : Clipping.typ;
  mutable stopping : Stopping.typ;
  mutable checkpoint : Checkpoint.typ;
  mutable verbosity : bool;
  }

Type definition of paramater.

val default : unit -> typ

Create module typ with default values.

val config : ?batch:Batch.typ -> ?gradient:Gradient.typ -> ?loss:Loss.typ -> ?learning_rate:Learning_Rate.typ -> ?regularisation:Regularisation.typ -> ?momentum:Momentum.typ -> ?clipping:Clipping.typ -> ?stopping:Stopping.typ -> ?checkpoint:Checkpoint.typ -> ?verbosity:bool -> float -> typ

This function creates a parameter object with many configurations.

source code

val to_string : typ -> string

Convert the module typ to its string representation.

Core functions

val minimise_weight : ?state:Checkpoint.state -> Params.typ -> (t -> t -> t) -> t -> t -> t -> Checkpoint.state * t

This function minimises the weight w of passed-in function f.

  • f is a function f : w -> x -> y.
  • w is a row vector but y can have any shape.

source code

val minimise_network : ?state:Checkpoint.state -> Params.typ -> (t -> t * t array array) -> (t -> t array array * t array array) -> (t array array -> 'a) -> (string -> unit) -> t -> t -> Checkpoint.state

This function is specifically designed for minimising the weights in a neural network of graph structure. In Owl’s earlier versions, the functions in the regression module were actually implemented using this function.

source code

val minimise_fun : ?state:Checkpoint.state -> Params.typ -> (t -> t) -> t -> Checkpoint.state * t

This function minimises f : x -> y w.r.t x.

x is an ndarray; and y is an scalar value.

source code