//

// This software is released under the MIT License.

// https://opensource.org/licenses/MIT

​

/* ===

ml5 Example

LSTM Generator example with p5.js

This uses a pre-trained model on a corpus of Virginia Woolf

For more models see: https://github.com/ml5js/ml5-data-and-training/tree/master/models/charRNN

=== */

​

let charRNN;

let textInput;

let lengthSlider;

let tempSlider;

let button;

let runningInference = false;

​

function setup() {

  noCanvas();

​

  // Create the LSTM Generator passing it the model directory

  charRNN = ml5.charRNN('https://raw.githubusercontent.com/ml5js/ml5-data-and-models/main/models/charRNN/woolf/', modelReady);

​

  // Grab the DOM elements

  textInput = select('#textInput');

  lengthSlider = select('#lenSlider');

  tempSlider = select('#tempSlider');

  button = select('#generate');

​

  // DOM element events

  button.mousePressed(generate);

  lengthSlider.input(updateSliders);

  tempSlider.input(updateSliders);

}

​

// Update the slider values

function updateSliders() {

  select('#length').html(lengthSlider.value());

  select('#temperature').html(tempSlider.value());

}

​

function modelReady() {

  select('#status').html('Model Loaded');

}

​

// Generate new text

function generate() {

  // prevent starting inference if we've already started another instance

  // TODO: is there better JS way of doing this?

  if(!runningInference) {

    runningInference = true;

​

    // Update the status log

    select('#status').html('Generating...');

​

    // Grab the original text

    const original = textInput.value();

    // Make it to lower case

    const txt = original.toLowerCase();

​

    // Check if there's something to send

    if (txt.length > 0) {

      // This is what the LSTM generator needs

      // Seed text, temperature, length to outputs

      // TODO: What are the defaults?

      const data = {

        seed: txt,

        temperature: tempSlider.value(),

        length: lengthSlider.value()

      };

​

      // Generate text with the charRNN

      charRNN.generate(data, gotData);

​

      // When it's done

      function gotData(err, result) {

        // Update the status log

        select('#status').html('Ready!');

        select('#result').html(txt + result.sample);

        runningInference = false;

      }

    }

  }

}

​