Unlock AI for Front‑End Engineers: Introducing Pipcook 1.0 and Its Powerful Features

Pipcook, an open‑source machine‑learning toolkit for front‑end engineers, was launched as version 1.0 after six months of development and contributions from over ten community members.

GitHub: https://github.com/alibaba/pipcook

Why AI for JavaScript?

Machine learning is rapidly advancing, but JavaScript ecosystems lag behind due to immature libraries and delayed support for cutting‑edge algorithms.

JavaScript’s ML ecosystem is not yet mature or stable.

Support for the latest deep‑learning models is often behind.

Pipcook addresses these issues by using Boa to bridge Python ecosystems, allowing most of the implementation to be in languages other than JavaScript—only about 25% of the codebase is TypeScript/JavaScript.

The core concept is the Pipeline , which defines a machine‑learning model in JSON and orchestrates data collection, access, processing, model definition, training, and evaluation.

What is a Pipeline?

A typical pipeline follows these steps:

Data collect plugin converts raw data to Pipcook’s standard format.

Data access plugin transforms the standard format into a UniDataset for the model.

Model define plugin creates the model architecture based on metadata.

Model train plugin trains the model with the training set.

Model evaluate plugin assesses the model with the test set.

Example pipeline configuration:

{
  "plugins": {
    "dataCollect": {
      "package": "@pipcook/plugins-mnist-data-collect",
      "params": {"trainCount": 8000, "testCount": 2000}
    },
    "dataAccess": {"package": "@pipcook/plugins-pascalvoc-data-access"},
    "dataProcess": {"package": "@pipcook/plugins-image-data-process", "params": {"resize": [224, 224]}},
    "modelDefine": {"package": "@pipcook/plugins-tensorflow-mobilenet-model-define", "params": {"batchSize": 8, "freeze": true}},
    "modelTrain": {"package": "@pipcook/plugins-image-classification-tensorflow-model-train", "params": {"epochs": 15}},
    "modelEvaluate": {"package": "@pipcook/plugins-image-classification-tensorflow-model-evaluate"}
  }
}

Run the pipeline with:

pipcook run https://foobar.xzz/mobilenet-classification.json

After training, the generated model and JavaScript library appear in the output directory.

$ cd ./output && npm install
$ node -e "require('./output')('input data')"

Pipcook also provides PipApp , an experimental feature that abstracts away training details, letting developers focus on business logic. Example PipApp code:

import { createLearnable, nlp } from '@pipcook/app';
const isCooking = createLearnable(async (sentence) => {
  return await nlp.classify(sentence);
});
const isBooking = createLearnable(async (sentence) => {
  return await nlp.classify(sentence);
});
(async () => {
  console.log(await isCooking('test'));
  console.log(await isBooking('booking test'));
})();

Plugin development follows a defined schema in package.json, specifying the plugin category and data type, and optionally Python dependencies via a conda field. Example plugin definition:

{
  "name": "my-own-pipcook-plugin",
  "version": "1.0.0",
  "description": "my own pipcook plugin",
  "dependencies": {"@pipcook/pipcook-core": "^0.5.0"},
  "pipcook": {"category": "dataCollect", "datatype": "image"},
  "conda": {"python": "3.7", "dependencies": {"tensorflow": "2.2.0"}}
}

Sample data‑collect plugin code:

const collectTextline: DataCollectType = async (args) => {
  const { uri, dataDir } = args;
  await fs.copy(uri, dataDir + '/my-own-dataset.csv');
  return null;
};
export default collectTextline;

The roadmap for future releases includes faster, more stable plugin installation, easier plugin development and sharing, more efficient pipeline execution, enhanced Pipboard usability, and integration of reinforcement learning to further empower front‑end AI workflows.