• namespace: Rindow\NeuralNetworks\Data\Dataset
• classname: NDArrayDataset

NDArray data preprocessor.

Read NDArray and process it on demand to generate a dataset.

Get the input NDArray by dividing it into batch size units. The selected data can be processed on demand using filters.

Also, the sufful option is specified by default, and it is selected from NDArray in a random order.

You can specify the data for tests and return the data as a set with inputs, or you can use a filter to generate a set from inputs.

Methods

constructor

$builer->NDArrayDataset(
        NDArray|array $inputs,
        NDArray $tests=null,
        int $batch_size=32,
        $shuffle=true,
        DatasetFilter $filter=null,
)

You can create a NDArrayDataset instances with the Data Builder.

Arguments

• tests: Specify inputs and the corresponding tests data.
• batch_size: batch size.
• shuffle: Randomize the order of the returned values.
• filter: Instance of dataset filter.

Examples

use Rindow\NeuralNetworks\Builder\NeuralNetworks;
$nn = new NeuralNetworks($mo);
$filter = new MyFilter();
$dataset = $nn->data()->NDArrayDataset($inputs,
    tests:$tests,
    filter:$filter);
foreach ($dataset as $batchdataset) {
    [$train,$label] = $batchdataset;
    foreach ($train as $key => $value) {
        $inputs = $value;
        $trues  = $label[$key];
        //....... some processing
    }
}

setFilter

public function setFilter(DatasetFilter $filter) : void

Set a filter to process input data on demand.

Arguments

• filter: Instance of conversion filter.

batchSize

public function batchSize() : int

Dataset batch size.

datasetSize

public function datasetSize() : int

Total size of the dataset.

In the initial state, the file is not read, so the correct size cannot be returned. You can get the size after reading to the end of the dataset

count

public function count() : int

Number of batch steps.

In the initial state, the file is not read, so the correct size cannot be returned. You can get the size after reading to the end of the dataset

How to make a filter

It is loaded by the fgetcsv function and passed a PHP native Array organized by batch size. You need to implement a filter that converts this to an NDArray.

Create a class that implements DatasetFilter and implement the translate method.

translate

public function translate(iterable $inputs, iterable $tests=null, $options=null) : array

Arguments

• inputs: NDArray rows of the specified batch size from the inputs.
• tests: NDArray rows of the specified batch size from the tests.
• options: N/A.

Output set

• inputs: Data expected to be used for training input
• tests: Data expected to be used for correct labels

Filter Example

An example of a filter that returns input data in 1/10

use Rindow\NeuralNetworks\Data\Dataset\DatasetFilter;

class TestFilter implements DatasetFilter
{
    public function __construct($mo = null)
    {
        $this->mo = $mo;
    }
    public function translate(
        iterable $inputs, iterable $tests=null, $options=null) : array
    {
        $la = $this->mo->la();
        $inputs = $la->copy($inputs);
        $la->scal(1/10,$inputs);
        return [$inputs,$tests];
    }
}