README

tensorflow-serving-php-client is a framework agnostic library helps you with calling machine learning models served by tensorflow serving from your php application. the library is inspired by tfserving-python-predict-client python client.

Requirements

tensorflow-serving-php-client requires grpc extension which you can install by running the command below:

sudo pecl install grpc

Installation

you can install the library via composer by running the following command inside your project root dir

composer require harranali/tensorflow-serving-php-client

Usage

an example in laravel

use Harranali\Tfserving\Client;
use Harranali\Tfserving\PredictMessage;
use Harranali\Tfserving\DataTypes;

Route::get('/', function () {
    $host = '192.168.99.100:9000';
    $modelName = 'my_model';
    $modelVersion = 1;
    $input = [12345];

    $tfsClient = new Client($host, $modelName, $modelVersion);

    $predictmessage = new PredictMessage;
    $predictmessage->setInputTensorName('a')
                   ->setInputTensorType(DataTypes::DT_INT32)
                   ->setInputTensorValue($input);

    $response = $tfsClient->predict($predictmessage);
    return $response;
});

First import the following classes:

use Harranali\Tfserving\Client;
use Harranali\Tfserving\PredictMessage;
use Harranali\Tfserving\DataTypes;

Second initiate the client by providing the parameters host:port, modelName, and modelVersion

$hostPort = 'localhost:9000';
$modelName = 'simple';
$modelVersion = 1;
$tfsClient = new Client($hostPort, $modelName, $modelVersion);

Third prepare the predict request message object

$predictmessage = new PredictMessage;
$predictmessage->setInputTensorName('a')
               ->setInputTensorType(DataTypes::DT_INT32)
               ->setInputTensorValue([1]);

here is a list of all data types available in the class DataTypes

DataTypes::DT_FLOAT
DataTypes::DT_DOUBLE
DataTypes::DT_INT32
DataTypes::DT_UINT8
DataTypes::DT_INT16
DataTypes::DT_INT8
DataTypes::DT_STRING
DataTypes::DT_COMPLEX64
DataTypes::DT_INT64
DataTypes::DT_BOOL
DataTypes::DT_QINT8
DataTypes::DT_QUINT8
DataTypes::DT_QINT32
DataTypes::DT_BFLOAT16
DataTypes::DT_QINT16
DataTypes::DT_QUINT16
DataTypes::DT_UINT16
DataTypes::DT_COMPLEX128
DataTypes::DT_HALF
DataTypes::DT_RESOURCE
DataTypes::DT_FLOAT_REF
DataTypes::DT_DOUBLE_REF
DataTypes::DT_INT32_REF
DataTypes::DT_UINT8_REF
DataTypes::DT_INT16_REF
DataTypes::DT_INT8_REF
DataTypes::DT_STRING_REF
DataTypes::DT_COMPLEX64_REF
DataTypes::DT_INT64_REF
DataTypes::DT_BOOL_REF
DataTypes::DT_QINT8_REF
DataTypes::DT_QUINT8_REF
DataTypes::DT_QINT32_REF
DataTypes::DT_BFLOAT16_REF
DataTypes::DT_QINT16_REF
DataTypes::DT_QUINT16_REF
DataTypes::DT_UINT16_REF
DataTypes::DT_COMPLEX128_REF
DataTypes::DT_HALF_REF
DataTypes::DT_RESOURCE_REF

Fourth Making the call to the server

$response = $tfsClient->predict($predictmessage);
return $response;

Contributing

all kind of contributions are welcomed.

License

tensorflow-serving-php-client is licensed under the MIT license.

harranali / tensorflow-serving-php-client

Maintainers

Details