JSONM

Description

JSONM is a module that creates Tensorflow models with JSON.

Jump right in

JSONM is designed so software engineers and machine learning engineers can quickly build, train, save, and load Tensorflow based AI & ML Models. The JSONM Library includes a JSONM UMD module with batteries included so you can use JSONM in the browser without transpilers or any additional setup/configuration. The JSONM UMD is ideal for JAMstack Applications.

Usage

The idea behind JSONM is to enable rapid model development. JSONM attempts to automate the data pre-processing and feature engineering needed for most modeling tasks.

Data Scientists who are more comfortable finely tuning hyperparameters and controlling pre-processing, scaling and normalization of datasets can also configure JSONM to meet specific model requirements.

What's included

JSONM currently supports the following models

• General purpose multivariate based predictions
• Multivariate linear regression based predictions
• General purpose multivariate classification based descriptions
• General purpose multivariate timeseries based forecasts
• General purpose content based recommendations (coming soon)
• General purpose cohort analysis (coming soon)

Installation

$ npm i @jsonstack/jsonm

JSONM Manual

• Getting Started
• Working With Data
  • Data Fetching
  • Feature Engineering
• Working With Models
  • Model Training
  • Model Evaluation
  • Model Predictions
  • Saving and Loading Models
• Advanced Topics
  • JSONM & JML Spec
  • Examples
  • Full API Docs

Basic Usage

import * as tf from '@tensorflow/tfjs-node';
import { getModel, setBackend, } from '@jsonstack/jsonm';

//set tensorflow
setBackend(tf);

//Iris Dataset e.g from https://raw.githubusercontent.com/repetere/modelx-model/master/src/test/mock/data/iris_data.csv
const type = 'ai-classification';
const dataset = [
  {
    "sepal_length_cm": 5.1,
    "sepal_width_cm": 3.5,
    "petal_length_cm": 1.4,
    "petal_width_cm": 0.2,
    "plant": "Iris-setosa",
  },
//  ...
  {
    "sepal_length_cm": 7.0,
    "sepal_width_cm": 3.2,
    "petal_length_cm": 4.7,
    "petal_width_cm": 1.4,
    "plant": "Iris-versicolor",
  },
  // ...
  {
    "sepal_length_cm": 5.9,
    "sepal_width_cm": 3.0,
    "petal_length_cm": 5.1,
    "petal_width_cm": 1.8,
    "plant": "virginica",
  }
]
const inputs = ['sepal_length_cm','sepal_width_cm','petal_length_cm','petal_width_cm', ];
const outputs = [ 'plant',];
const on_progress = ({ completion_percentage, loss, epoch, status, logs, defaultLog, }) => { 
  console.log({ completion_percentage, loss, epoch, status, logs, defaultLog, });
}
const IrisModel = await getModel({
  type,
  dataset,
  inputs,
  outputs,
  on_progress,
}); 
await IrisModel.trainModel()
const predictions = await IrisModel.predictModel({ 
  prediction_inputs:[
    { sepal_length_cm: 5.1, sepal_width_cm: 3.5, petal_length_cm: 1.4, petal_width_cm: 0.2, },
    { sepal_length_cm: 5.9, sepal_width_cm: 3.0, petal_length_cm: 5.1, petal_width_cm: 1.8, },
  ],
}); // => [ { plant:'Iris-setosa' }, { plant:'Iris-virginica' }, ]

Example

Development

Note Make sure you have typescript installed

$ npm i -g typescript 

For generating documentation

$ npm run doc

Notes

Check out https://repetere.github.io/jsonm/ for the full jsonm Documentation

Testing

$ npm test

Contributing

Fork, write tests and create a pull request!

License

MIT