Google AutoML Prediction with a Google Cloud Storage source

As per the Gist


// --------------------------------------------------
//   Example Google Cloud AutoML Prediction
// --------------------------------------------------
// @author Michael Kubler
// @date 2020-10-07th
// This is a cut down gist of what you need to
// make a Google Cloud AutoML (Auto Machine Learning)
// prediction request, based off an already uploaded
// file in Google Cloud Storage (GCS).
// The main point is that the payload to be provided
// needs to include a Document
// the Document needs to have an DocumentInputConfig
// The DocumentInputConfig needs a GcsSource
// Those things took longer than they should have to
// find and work out how to use.
// The Documentation is auto-generated and hard to
// understand.
// Semi-Useful links:

use Google\Cloud\AutoMl\V1\PredictionServiceClient;
use Google\Cloud\AutoMl\V1\AnnotationPayload;
use Google\Cloud\AutoMl\V1\Document;
use Google\Cloud\AutoMl\V1\DocumentInputConfig;
use Google\Cloud\AutoMl\V1\ExamplePayload;
use Google\Cloud\AutoMl\V1\GcsSource;
use yii\helpers\VarDumper;

// -- Things to change
$autoMlProject = '186655544321'; // The ProjectId - Set this to your own
$autoMlLocation = 'us-central1'; // For AutoML this is likely to be the location
$autoMlModelId = 'TEN15667778886635554442'; // The modelId - Set this to your own
$autoMlCredentialsLocation = __DIR__ . '/google-service-account.json'; // Set this to where ever you set your auth credentials file
$gsFilePath = 'gs://<bucket>/filePath.pdf'; // Obviously set this to your file location in Google Cloud Storage

// -- General setup
putenv('GOOGLE_APPLICATION_CREDENTIALS=' . $autoMlCredentialsLocation);
$autoMlPredictionServiceClient = new PredictionServiceClient();
$autoMlPredictionServiceFormattedParent = $autoMlPredictionServiceClient->modelName($autoMlProject, $autoMlLocation, $autoMlModelId);

// -- Setup the request
$pdfGsLocation = (new GcsSource())->setInputUris([$gsFilePath]);
$pdfDocumentConfig = (new DocumentInputConfig())->setGcsSource($pdfGsLocation);
$pdfDocument = (new Document())->setInputConfig($pdfDocumentConfig);
$payload = (new ExamplePayload())->setDocument($pdfDocument);

// -- Make the request (Here we actually do the prediction)
$autoMlFullResponse = $autoMlPredictionServiceClient->predict($autoMlPredictionServiceFormattedParent, $payload);

// --------------------------------------------------
//   Output #1 - All as JSON
// --------------------------------------------------
// You've got a couple of options now, you could return the full set by outputting / returning the serializeToJsonString response
echo $autoMlFullResponse->serializeToJsonString();

// --------------------------------------------------
//   Output #2 - Get just specific fields
// --------------------------------------------------
// Or for this example you might only want the payload[i].displayName and payload[i].textExtraction.textSegment.content
$payload = $autoMlFullResponse->getPayload();
$autoMlProcessedResponse = [];
foreach ($payload->getIterator() as $payloadEntry) {
    /** @var AnnotationPayload $payloadEntry */
    $autoMlProcessedResponse[$payloadEntry->getDisplayName()] = $payloadEntry->getTextExtraction()->getTextSegment()->getContent();
echo VarDumper::export($autoMlProcessedResponse); // PHP array format, you'd probably want to JSON encode it instead

// NB: You'll likely want to convert this to a class and provide the $gsFilePath in a method and return the expected response not output it

Published by

Michael Kubler

Photographer, cinematographer, web master/coder.

Leave a Reply

Your email address will not be published. Required fields are marked *