Make (owl-base.Owl_neural

Parameters

module E : Owl_types_computation_engine.Sig

Signature

module Engine : sig ... end

module Neural : sig ... end

Naive compilation functions, need to pass in loss function

val compile_simple : 
  Neural.Graph.network ->
  int array ->
  (Neural.Algodiff.t ->
    Neural.Graph.Neuron.Optimise.Algodiff.t ->
    Neural.Graph.Neuron.Optimise.Algodiff.t) ->
  Neural.Algodiff.A.arr
  * Neural.Algodiff.A.arr
  * Neural.Algodiff.A.arr array
  * Neural.Algodiff.A.arr array

Shallow compilation functions, includes only gradient

val compile_shallow : 
  Neural.Params.typ ->
  Neural.Graph.network ->
  int ->
  Neural.Algodiff.t
  * Neural.Algodiff.t
  * Neural.Graph.Neuron.Optimise.Algodiff.t array
  * Neural.Graph.Neuron.Optimise.Algodiff.t array
  * Neural.Graph.Neuron.Optimise.Algodiff.t

Deep compilation functions, includes gs, us, ps, ch, and new weights

val compile_deep : 
  Neural.Params.typ ->
  Neural.Graph.network ->
  int ->
  Neural.Graph.Neuron.Optimise.Algodiff.t
  * Neural.Algodiff.t
  * Neural.Algodiff.t
  * Engine.graph

val make_eval_fun : 
  'a ->
  Neural.Algodiff.t ->
  Neural.Algodiff.t ->
  Engine.Graph.graph ->
  Neural.Algodiff.t ->
  Neural.Algodiff.t ->
  'b

val make_update_fun : Engine.graph -> unit -> unit

val train : 
  ?state:Neural.Optimise.Checkpoint.state ->
  ?params:Neural.Params.typ ->
  Neural.Graph.network ->
  Neural.Algodiff.t ->
  Neural.Optimise.Algodiff.t ->
  Neural.Optimise.Checkpoint.state

val model_inputs : 
  ?optimise:bool ->
  ?batch_size:int ->
  Neural.Graph.network ->
  Neural.Algodiff.t array ->
  Neural.Algodiff.t array

val model : 
  ?optimise:bool ->
  ?batch_size:int ->
  Neural.Graph.network ->
  Neural.Algodiff.t ->
  Neural.Algodiff.t