ryanrhymes

#require "owl_neural";;
open Owl_neural;;

(* config the neural network *)
let nn = Feedforward.create ();;
let l0 = linear 784 300 ~init_typ:Init.(Uniform (-0.075,0.075));;
let l1 = linear 300 10 ~init_typ:Init.(Uniform (-0.075,0.075));;
Feedforward.add_layer nn l0 ~act_typ:Activation.Tanh;;
Feedforward.add_layer nn l1 ~act_typ:Activation.Softmax;;
print nn;;

ryanrhymes

#require "owl_neural";;
open Owl_neural;;

(* config the neural network *)
let nn = Feedforward.create ();;
let l0 = linear ~inputs:784 ~outputs:300 ~init_typ:Init.(Uniform (-0.075,0.075));;
let l1 = linear ~inputs:300 ~outputs:10 ~init_typ:Init.(Uniform (-0.075,0.075));;
Feedforward.add_layer nn l0;;
Feedforward.add_activation nn Activation.Tanh;;
Feedforward.add_layer nn l1;;

ryanrhymes

#require "owl_neural";;
open Owl_neural;;

let nn = Feedforward.create ();;
let l0 = linear ~inputs:784 ~outputs:300 ~init_typ:(Init.Uniform (-0.075,0.075));;
let l1 = linear ~inputs:300 ~outputs:10 ~init_typ:(Init.Uniform (-0.075,0.075));;
Feedforward.add_layer nn l0;;
Feedforward.add_activation nn Activation.Tanh;;
Feedforward.add_layer nn l1;;
Feedforward.add_activation nn Activation.Softmax;;

ryanrhymes

(* Owl Lib: https://github.com/ryanrhymes/owl *)

open Owl_ext;;

let a = F 5.;;                             (* float number     *)
let b = C Complex.({re = 2.; im = 5.});;   (* complex number   *)
let x = Dense.Matrix.S.uniform 3 3;;       (* float32 matrix   *)
let y = Dense.Matrix.D.uniform 3 3;;       (* float64 matrix   *)
let z = Dense.Matrix.Z.uniform 3 3;;       (* complex64 matrix *)

ryanrhymes

let test_model nn x y =
  Mat.iter2_rows (fun u v ->
    Dataset.print_mnist_image (unpack_mat u);
    let p = run_network u nn |> unpack_mat in
    Owl.Mat.print p;
    Printf.printf "prediction: %i\n" (let _, _, j = Owl.Mat.max_i p in j)
  ) x y

let _ =
  let x, _, y = Dataset.load_mnist_train_data () in

ryanrhymes

let backprop nn eta x y =
  let t = tag () in
  Array.iter (fun l ->
    l.w <- make_reverse l.w t;
    l.b <- make_reverse l.b t;
  ) nn.layers;
  let loss = Maths.(cross_entropy y (run_network x nn) / (F (Mat.row_num x |> float_of_int))) in
  reverse_prop (F 1.) loss;
  Array.iter (fun l ->
    l.w <- Maths.((primal l.w) - (eta * (adjval l.w))) |> primal;

ryanrhymes

open Owl
open Algodiff.AD

type layer = { mutable w : t; mutable b : t; a : t -> t }
type network = { layers : layer array }

let run_layer x l = Maths.((x $@ l.w) + l.b) |> l.a
let run_network x nn = Array.fold_left run_layer x nn.layers

let l0 = {

ryanrhymes

open Owl_algodiff_ad

type layer = { mutable w : t; mutable b : t; a : t -> t }

type network = { layers : layer array }

let run_layer x l = Maths.((x $@ l.w) + l.b) |> l.a

let run_network x nn = Array.fold_left run_layer x nn.layers

ryanrhymes

(* Test neural network on MNIST *)

#require "bitstring";;
#require "bitstring.syntax";;

open Bitstring

type t = { magic : int; items : int }

let dataset = "/Users/liang/owl_dataset/t10k-images-idx3-ubyte"

ryanrhymes

(*
 * Eigen - an OCaml interface to C++ Eigen library
 * Copyright (c) 2016 Liang Wang <liang.wang@cl.cam.ac.uk>
 *)

module type MatSig = sig

  type elt
  type mat