php-ml linear classifier performance (partial training vs full training in 1 batch) (https://github.com/php-ai/php-ml/pull/78)

gistfile1.txt

RESULTS:
➜  php-ml git:(partial-train) ✗ clear ; php partial-vs-full.php
Classifier perceptron full train memory usage: 
  Usage from before data reading to end of training: 231704
Classifier adaline full train memory usage: 
  Usage from before data reading to end of training: 265160
Classifier logistic full train memory usage: 
  Usage from before data reading to end of training: 248408
Classifier perceptron partial train memory usage: 
  Usage from before data reading to end of training: 9176
Classifier adaline partial train memory usage: 
  Usage from before data reading to end of training: 86136
Classifier logistic partial train memory usage: 
  Usage from before data reading to end of training: 17336


SCRIPT (checkout partial-train branch)
<?php

require __DIR__ . '/vendor/autoload.php';

use Phpml\Classification\Linear\Perceptron;
use Phpml\Classification\Linear\Adaline;
use Phpml\Classification\Linear\LogisticRegression;

define('BATCH_SIZE', 10);

$classifiers['perceptron'] = new Perceptron();
$classifiers['adaline'] = new Adaline();
$classifiers['logistic'] = new LogisticRegression();

foreach ($classifiers as $name => $classifier) {
    fullRun($name, $classifier);
}

$classifiers['perceptron'] = new Perceptron();
$classifiers['adaline'] = new Adaline();
$classifiers['logistic'] = new LogisticRegression();

foreach ($classifiers as $name => $classifier) {
    partialRun($name, $classifier);
}

function fullRun($name, $classifier) {
    $filepath = __DIR__ . '/data/wine.csv';
    $handle = fopen($filepath, 'rb');
    $dontCareAboutHeaders = fgetcsv($handle, 1000);
    echo 'Classifier ' . $name . ' full train memory usage: ' . PHP_EOL;
    $membefore = memory_get_usage(false);

    $samples = [];
    $targets = [];
    $features = null;
    while (($data = fgetcsv($handle, 1000)) !== false) {
        if ($features == null) {
            $features = count($data) - 1;
        }

        $samples[] = array_slice($data, 0, $features);
        $targets[] = $data[$features];
    }
    fclose($handle);

    $mem1 = memory_get_usage(false);
    echo '  Usage during data reading: ' . ($mem1 - $membefore) . PHP_EOL;
    $classifier->train($samples, $targets);
    $mem2 = memory_get_usage(false);
    echo '  Usage during training: ' . ($mem2 - $mem1) . PHP_EOL;
    echo '  Usage from before data reading to end of training: ' . ($mem2 - $membefore) . PHP_EOL;
}


function partialRun($name, $classifier) {
    $filepath = __DIR__ . '/data/wine.csv';
    $handle = fopen($filepath, 'rb');
    $dontCareAboutHeaders = fgetcsv($handle, 1000);
    echo 'Classifier ' . $name . ' partial train memory usage: ' . PHP_EOL;
    $membefore = memory_get_usage(false);

    $samples = [];
    $targets = [];
    $features = null;
    while (($data = fgetcsv($handle, 1000)) !== false) {
        if ($features == null) {
            $features = count($data) - 1;
        }

        $samples[] = array_slice($data, 0, $features);
        $targets[] = $data[$features];

        if (count($samples) === BATCH_SIZE) {

            $mem1 = memory_get_usage(false);
            echo '  Usage reading data batch: ' . ($mem1 - $membefore) . PHP_EOL;
            $classifier->partialTrain($samples, $targets, array(1, 2, 3));
            $mem2 = memory_get_usage(false);
            echo '  Usage during partial training: ' . ($mem2 - $mem1) . PHP_EOL;

            $samples = [];
            $targets = [];
        }
    }
    fclose($handle);
    echo '  Usage from before data reading to end of training: ' . (memory_get_usage(false) - $membefore) . PHP_EOL;
}