Skip to content

Latest commit

 

History

History
277 lines (203 loc) · 6.94 KB

README.md

File metadata and controls

277 lines (203 loc) · 6.94 KB
Raw

deepdetect-js

All Contributors

DeepDetect JS client

Files

  • src/index.js - client source code

  • src/index.test.js - client methods tests

  • doc/web-example/server.js - simple webserver to serve web-example index.html and proxy api calls to a deepdetect server

  • doc/web-example/index.html - deepdetect-js web integration demo

Usage

Web integration

DeepDetect-JS can be used on a webpage, you probably should run deepdetect server behind a http-proxy to avoid same-origin policy issues.

A simple webserver demo is available on http://localhost:3000 when running the following command:

yarn run web-example

Here is the simple /info api call on a DeepDetect server. Note the {path: 'api'} parameter when initializing DD object.

...
<script src="https://cdn.jsdelivr.net/npm/deepdetect-js@0.0.0-development/dist/deepdetect-browser.min.js"></script>
<script>
  async function fetchInfo() {
    const dd = new deepdetect.DD({path: 'api'});
    const info = await dd.info();
    document.getElementById('infoResult').innerHTML = JSON.stringify(info);
  }

  fetchInfo();
</script>
...

NodeJS integration

Following usage examples will use nodejs, install it with this command:

npm install --save deepdetect-js

Connect to DeepDetect server, and fetch informations

Here is the simplest way to get information about a DeepDetect server:

import DD from 'deepdetect-js';

async () => {

  const dd = new DD()

  // Get DeepDetect server info
  const info = await dd.info()
  console.log(info);

}

You can also specified the DeepDetect server host and port options:

import DD from 'deepdetect-js';

async () => {

  const dd = new DD('10.10.10.1', 8580)

  // Get DeepDetect server info
  const info = await dd.info()
  console.log(info);

}

Service API

Once connected to a DeepDetect server, the Service API allows to:

  • create a service
  • fetch informations about a service
  • delete a service
import DD from 'deepdetect-js';

async () => {

  const dd = new DD()

  // Create a service
  const serviceName = 'myserv';

  const serviceConfig = {
    description: 'example classification service',
    model: {
      repository: '/home/me/models/example',
      templates: '../templates/caffe'
    },
    mllib: 'caffe',
    parameters: {
      input: { connector: 'txt' },
      mllib: { nclasses: 20 },
      output: {},
    },
  };

  const createService = await dd.putService(serviceName, serviceConfig)

  // Fetch service information
  const service = await dd.getService(serviceName);
  console.log(service);

  // Delete service
  const deleteService = await dd.deleteService(serviceName, {clear: 'full'});
}

Train API

Once connected to a DeepDetect server, the Train API allows to:

  • Create a training job
  • Get information on a non-blocking training job
  • Kills a non-blocking training job
import DD from 'deepdetect-js';

async () => {

  const dd = new DD()
  const serviceName = 'myserv';

  // Create a training job
  const train = await dd.postTrain(
    serviceName,
    [ '/home/me/deepdetect/examples/all/n20/news20' ],
    {
      test_split: 0.2,
      shuffle: true,
      min_count: 10,
      min_word_length: 3,
      count: false,
    },
    {
      gpu: false,
      solver: {
        iterations: iterationsN20,
        test_interval: 200,
        base_lr: 0.05,
        snapshot: 2000,
        test_initialization: true,
      },
      net: {
        batch_size: 100,
      },
    },
    { measure: ['acc', 'mcll', 'f1'] },
    false
  );

  // Get information on a non-blocking training job
  const trainingJob = await dd.getTrain(serviceName);
  console.log(trainingJob);

  // Kills a non-blocking training job
  const deletedTrainingJob = await dd.deleteTrain(serviceName);
  console.log(deletedTrainingJob);

}

Predict API

Once connected to a DeepDetect server, the Predict API allows to makes prediction from data and model

import DD from 'deepdetect-js';

async () => {

  const dd = new DD()
  const serviceName = 'myserv';

  // Predict with measures
  const postData = {
    service: serviceName,
    data: [ '/home/me/deepdetect/examples/all/n20/news20' ],
    parameters: {
      input: {},
      mllib: {
        gpu: false,
        net: {
          test_batch_size: 10,
        },
      },
      output: {
        measure: ['f1']
      }
    }
  };

  const predict = await dd.postPredict(postData)
  console.log(predict);

}

Build and release

  1. Modify version number in package.json
  2. npm run build
  3. npm publish - documentation

Testing

In order to run the test, you first need to run a deepdetect server loccaly on port 8080. To do so, you can use the following docker command:

docker run -d -p 8080:8080 docker.jolibrain.com/deepdetect_cpu

Then you can run the test suite:

yarn test

If you find and issue with your tests, please check the header parameters available in src/index.test.js.

Changelog

  • 1.8.12 - 21/03/2024 - return !response.ok error from http requests
  • 1.8.11 - 19/10/2023 - Add missing process lib
  • 1.8.10 - 19/10/2023 - Review dependabot alerts
  • 1.8.9 - 17/10/2023 - Update dependencies
  • 1.8.8 - 19/10/2021 - Add option to enable/disable Accept-Encoding gzip request header
  • 1.8.7 - 05/01/2020 - Replace NaN values in returned json from deepdetect server
  • 1.8.4 - 16/10/2020 - Fix conditional check of options.sameOrigin by eh-dub

Contributors

Thanks goes to these people (emoji key):


Alexandre Girard
💻
Ariel Weingarten
💻

This project follows the all-contributors specification. Contributions of any kind welcome!

License

MIT © Jolibrain