test_mnist_03.ml

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
  for i = 1 to 500 do
    let x', y' = Dataset.draw_samples x y 100 in
    backprop nn (F 0.01) (Mat x') (Mat y')
    |> Printf.printf "#%i : loss=%g\n" i
    |> flush_all;
  done;
  let x, y, _ = Dataset.load_mnist_test_data () in
  let x, y = Dataset.draw_samples x y 10 in
  test_model nn (Mat x) (Mat y)