Features and labels

final class Featurizer
{
    private array $vocab = [];   // for each word column: every value we've seen
    private array $min = [], $max = [], $mean = [], $std = [];

    public function __construct(private array $numeric, private array $categorical) {}

    // Learn what's needed to encode: the list of possible words, and the
    // min/max/mean/spread of every column (needed for scaling).
    public function fit(array $rows): void
    {
        foreach ($this->categorical as $col) {
            $values = array_values(array_unique(array_column($rows, $col)));
            sort($values);
            $this->vocab[$col] = $values;
        }
        $vectors = array_map(fn($r) => $this->encode($r), $rows);
        foreach (array_keys($this->names()) as $j) {
            $column = array_column($vectors, $j);
            $this->min[$j]  = min($column);
            $this->max[$j]  = max($column);
            $this->mean[$j] = array_sum($column) / count($column);
            $var = 0.0;
            foreach ($column as $v) $var += ($v - $this->mean[$j]) ** 2;
            $this->std[$j] = sqrt($var / count($column)) ?: 1.0;
        }
    }

    // Names of the columns in the vector, e.g. "weight", "color=red"
    public function names(): array
    {
        $names = $this->numeric;
        foreach ($this->categorical as $col) {
            foreach ($this->vocab[$col] as $value) $names[] = "$col=$value";
        }
        return $names;
    }

    // ONE-HOT ENCODING: a word column becomes several 0/1 columns,
    // one per possible word. "orange" → color=orange: 1, every other colour: 0.
    // (Why not red=1, orange=2, yellow=3? Because then yellow would be
    //  "more" than red, and red–yellow would be "farther" than red–orange.)
    public function encode(array $row): array
    {
        $v = [];
        foreach ($this->numeric as $col) $v[] = (float) $row[$col];
        foreach ($this->categorical as $col) {
            foreach ($this->vocab[$col] as $value) $v[] = $row[$col] === $value ? 1.0 : 0.0;
        }
        return $v;
    }

    // SCALING: put every column on a similar range so no column dominates.
    public function scale(array $v, string $how): array
    {
        foreach ($v as $j => $x) {
            if ($how === 'minmax') {            // squeeze into 0..1
                $range = $this->max[$j] - $this->min[$j];
                $v[$j] = $range ? ($x - $this->min[$j]) / $range : 0.0;
            } elseif ($how === 'standard') {    // "how many spreads from average?"
                $v[$j] = ($x - $this->mean[$j]) / $this->std[$j];
            }
        }
        return $v;
    }
}

function euclidean(array $a, array $b): float
{
    $sum = 0.0;
    foreach ($a as $j => $x) $sum += ($x - $b[$j]) ** 2;
    return sqrt($sum);
}