Skip to content

Latest commit

 

History

History
47 lines (39 loc) · 3.1 KB

README.md

File metadata and controls

47 lines (39 loc) · 3.1 KB
Raw

Interactive data generation and fitting

Background

See the codelab TensorFlow.js — Making Predictions from 2D Data for an introduction to TenosrFlow.js and also to the use-case and the code here.

Features

  • Interactive generation of (parameterized) test data
  • Interactive training and validation of (parameterized) Neural Network
  • Ease of extension of interactive parameters (sliders) thanks to a high level of abstraction and automation
  • Encapulation, reusability, and separation of concerns with a purely class-based code
    • Immutable implementation of logic (but not data models)

Use-Case

  • Samples of a decaying exponential function (with controllable sampling rate, noise, sparsity ...) are defined as x and y vectors.
  • An Artificial Neural Network is trained to learn the dependency y=f(x) with controllable learning parameters.
  • The plots x vs. y and x vs. f(x) are superimposed for a semi-quantitative assessment of the learning.

Execution

  • Browser client
    • Double-click on the file index.html (or open it with your browser of choice). No server is needed.
  • Node.js client
    • Run the command npm install and then npm test.
    • An example of this client is the GitHub Action of this project.
  • Deno client
    • Run the command Deno test ./tests/javascript/deno/test.js --allow-read.
    • An example of this client is the GitHub Action of this project.
      • If the main client is a Deno-client, it is recommended that the main code be imported in the test file. The current implementation is based on the restriction to support 3 different clients.

CI/CD

Snapshot of Control-Parameters and Outputs

The sliders for epochs (250 vs. 400) and noise (2.5% vs. 5%) provide a dynamic and user-defined compromise between accuracy and speed (Note that these stochastic results are reproducible only in a statistical sense).

Defining new sliders

In the function Controller.createSliders, simply extend the local variable sliders. For example to include a slider for sparsity, the definition can be extended as follows:

const sliders = {
  noisePercent: Controller.createSlider('Noise', 0, inputDataParameters.noisePercent * 2, sliderIndex++),
  epochs: Controller.createSlider('Epoch', 100, inputDataParameters.epochs * 2, sliderIndex++),
  sparsity: Controller.createSlider('Sparsity', 1, 6, sliderIndex++)
};

All other steps, including the setting of callbacks, will be done automatically.