grid-search

1.1.1 • Public • Published

Grid Search Icon

Grid Search

Build Status Codecov Status

grid-search is a small, simple node module that can be used to generate an array of parameters to use during a machine learning grid search. It will generate every possible combination of parameters based on user input.

This package additionally ships with a small range method that can be used to generate a range of values between two numbers.

Installation

Install grid-search using npm.

npm i grid-search

Example Use

Basic grid search parameter generation

const { gridSearch } = require("grid-search");
 
const params = {
  iterations: [10, 20, 30],
  objective: "binaryCrossentropy",
  dropout: [0.4, 0.5, 0.6]
};
 
const search = gridSearch(params);
 
console.log(search);

Your output will be:

[
  { iterations: 10, objective: "binaryCrossentropy", dropout: 0.4 },
  { iterations: 10, objective: "binaryCrossentropy", dropout: 0.5 },
  { iterations: 10, objective: "binaryCrossentropy", dropout: 0.6 },
  { iterations: 20, objective: "binaryCrossentropy", dropout: 0.4 },
  { iterations: 20, objective: "binaryCrossentropy", dropout: 0.5 },
  { iterations: 20, objective: "binaryCrossentropy", dropout: 0.6 },
  { iterations: 30, objective: "binaryCrossentropy", dropout: 0.4 },
  { iterations: 30, objective: "binaryCrossentropy", dropout: 0.5 },
  { iterations: 30, objective: "binaryCrossentropy", dropout: 0.6 }
];

Using range

The previous example can use the range function to specify the iterations and dropout parameters and achieve the exact same result.

range(start, finish, step)

const { range } = require("grid-search");
 
const params = {
  iterations: range(10, 30, 10),
  objective: "binaryCrossentropy",
  dropout: range(0.4, 0.6, 0.1)
};

Contributing

Contributions welcome! Please open an issue in the Github repository describing what changes you would like to see (or to contribute yourself).

Package Sidebar

Install

npm i grid-search

Weekly Downloads

2

Version

1.1.1

License

MIT

Unpacked Size

5.93 kB

Total Files

7

Last publish

Collaborators

  • nas5w