Last active
February 12, 2018 02:12
-
-
Save y-uti/1e04a3dc871a5287ccbea3283b4a3d9c to your computer and use it in GitHub Desktop.
A sample code calculating precision-recall curve from the result of the probability estimation by SVC
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
<?php | |
require_once __DIR__ . '/vendor/autoload.php'; | |
use Phpml\Classification\SVC; | |
use Phpml\SupportVectorMachine\Kernel; | |
// Load an example dataset published at the web site below | |
// https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html | |
// | |
// Note that this function can't be applied to other datasets because the file | |
// format is assumed not to be "sparse", see: | |
// https://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html#/Q03:_Data_preparation | |
function load_dataset($url) | |
{ | |
$samples = []; | |
$labels = []; | |
foreach (file($url, FILE_IGNORE_NEW_LINES) as $line) { | |
$columns = explode(' ', $line); | |
$labels[] = array_shift($columns); | |
$samples[] = array_map(function ($c) { return (float) explode(':', $c, 2)[1]; }, $columns); | |
} | |
return [$samples, $labels]; | |
} | |
$train_data_url = 'https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/svmguide1'; | |
[$train_samples, $train_labels] = load_dataset($train_data_url); | |
$test_data_url = 'https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary/svmguide1.t'; | |
[$test_samples, $test_labels] = load_dataset($test_data_url); | |
// Build an SVC instance then train and predict samples, with probability estimation enabled | |
$svc = new SVC(Kernel::RBF, 1.0, 3, 1, 0.0, 0.001, 100, true, true); | |
$svc->train($train_samples, $train_labels); | |
$probabilities = $svc->predictProbability($test_samples); | |
// Sort actual labels of the test set by descending order of probabilities for positive | |
$positiveness = array_map(function ($p) { return $p['1']; }, $probabilities); | |
$actuals = array_map(null, $positiveness, $test_labels); | |
usort($actuals, function ($a1, $a2) { return $a2[0] <=> $a1[0]; }); | |
$actuals = array_map(function ($a) { return $a[1]; }, $actuals); | |
// Calculate precision and recall | |
$n_positives = count(array_filter($actuals, function ($a) { return $a === '1'; })); | |
for ($k = 0, $tp = 0; $k < count($actuals); ++$k) { | |
$tp += $actuals[$k] === '1' ? 1 : 0; | |
$precision = $tp / ($k + 1); | |
$recall = $tp / $n_positives; | |
echo $precision, ',', $recall, PHP_EOL; | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Instead of
usort
on the line 43,rsort($actuals)
is also OK. (I always fail to choose a proper sorting method in PHP)