J. Roger Zhao jzstark

Here lists the examples of using tfgraph to convert computations in Owl to be executed on TensorFlow.

Files:

config.ml: MirageOS configuration file.
simple_mnist.ml: main logic of MNIST neural network.
simple_mnist_weight.ml: pre-trained weights of the neural network.

Information:

Weight file size: 146KB.
Model test accuracy: 92%.
MirageOS: compile with Unix backend. The generated binary is 10MB. The other backends are not tested yet.

How I think Algodiff Works

To train a network, we need to first use train_generic:

let train_generic ?state ?params ?(init_model=true) nn x y =
    if init_model = true then init nn;
    let f = forward nn in
    let b = backward nn in
    let u = update nn in

Files:

simple_mnist.ml: main logic of MNIST neural network.
simple_mnist_weight.ml: pre-trained weights of the neural network.

Information:

Weight file size: 146KB.
Model test accuracy: 92%.
MirageOS: compile with Unix backend. The generated binary is 10MB. The other backends are not tested yet. Here is the MirageOS version of this simple MNIST.

Usage:

Owl-Tensorflow Converter Example: MNIST CNN Training

This example is a example of MNIST-based CNN.

Step 1 : running OCaml script tfgraph_train.ml, which generates a file tf_convert_mnist.pbtxt in current directory.
Step 2 : make sure tf_convert_mnist.pbtxt and tfgraph_train.py in the same graph; make sure Tensorflow/numpy etc. is installed.
Step 3 : execute python tf_converter_mnist.py, and the expected output on screen is the training progress. After each 100 steps, loss value and model accuracy will be shown.

Here we only assume the python script writer knows where to find the output node (in collection "result") and the placeholder names (x:0).

Owl-Tensorflow Converter Example: MNIST CNN

This example is a example of MNIST-based CNN.

Step 1 : running OCaml script tfgraph_inf.ml, which generates a file tf_convert_mnist.pbtxt in current directory.
Step 2 : make sure tf_convert_mnist.pbtxt and tfgraph_inf.py in the same graph; make sure Tensorflow/numpy etc. is installed.
Step 3 : execute python tf_converter_mnist.py, and the expected output in screen is an array of size [100], each element is a boolean value. There is also an value to indicate the inference accuracy.

Here we only assume the python script writer knows where to find the output node (in collection "result") and the placeholder names (x:0).

	#!/usr/bin/python

	import numpy as np
	import random

	# first condition: average: 4.5
	def crossed1(p1, p2):
	x1, y1 = p1
	x2, y2 = p2
	# p1 should be within;

	open Owl
	open Algodiff.D


	let rec desc ?(eta=F 0.01) ?(eps=1e-6) f x =
	let g = (diff f) x in
	if (unpack_flt g) < eps then x
	else desc ~eta ~eps f Maths.(x - eta * g)


	<!DOCTYPE html>
	<html>
	<head>
	<meta charset="UTF-8">
	<title>Owl - OCaml Scientic and Engineering Computing</title>

	<script src="https://code.jquery.com/jquery-3.4.1.slim.min.js" integrity="sha384-J6qa4849blE2+poT4WnyKhv5vZF5SrPo0iEjwBvKU7imGFAV0wwj1yYfoRSJoZ+n" crossorigin="anonymous"></script>
	<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/umd/popper.min.js" integrity="sha384-Q6E9RHvbIyZFJoft+2mJbHaEWldlvI9IOYy5n3zV9zzTtmI3UksdQRVvoxMfooAo" crossorigin="anonymous"></script>
	<script src="https://stackpath.bootstrapcdn.com/bootstrap/4.4.1/js/bootstrap.min.js" integrity="sha384-wfSDF2E50Y2D1uUdj0O3uMBJnjuUD4Ih7YwaYd1iqfktj0Uod8GCExl3Og8ifwB6" crossorigin="anonymous"></script>

	open Printf

	let () =
	for i = 0 to Array.length Sys.argv - 1 do
	printf "[%i] %s\n" i Sys.argv.(i)
	done;;