MNIST Neural Network in PHP

This source code seeks to replicate the (now removed) MNIST For ML Beginners tutorial from the Tensorflow website using straight forward PHP code. Hopefully, this example will make that tutorial a bit more manageable for PHP developers.

The task is to recognise digits, such as the ones below, as accurately as possible.

By AndrewCarterUK

Contents

• mnist.php: Glue code that runs the algorithm steps and reports algorithm accuracy
• Dataset.php: Dataset container object
• DatasetReader.php: Retrieves images and labels from the MNIST dataset
• NeuralNetwork.php: Implements training and prediction routines for a simple neural network

Usage

php mnist.php

Description

The neural network implemented has one output layer and no hidden layers. Softmax activation is used, and this ensures that the output activations form a probability vector corresponding to each label. The cross entropy is used as a loss function.

This algorithm can achieve an accuracy of around 92% (with a batch size of 100 and 1000 training steps). That said, you are likely to get bored well before that point with PHP.

Expected Output

Loading training dataset... (may take a while)
Loading test dataset... (may take a while)
Starting training...
Step 0001	Average Loss 4.12	Accuracy: 0.19
Step 0002	Average Loss 3.21	Accuracy: 0.23
Step 0003	Average Loss 2.59	Accuracy: 0.32
Step 0004	Average Loss 2.43	Accuracy: 0.36
Step 0005	Average Loss 1.87	Accuracy: 0.45
Step 0006	Average Loss 2.06	Accuracy: 0.47
Step 0007	Average Loss 1.67	Accuracy: 0.51
Step 0008	Average Loss 1.81	Accuracy: 0.46
Step 0009	Average Loss 1.74	Accuracy: 0.55
Step 0010	Average Loss 1.24	Accuracy: 0.56
...