Function parseArguments

date-conversion-attention/train.js:127–175

()

}

function parseArguments() {
  const argParser = new argparse.ArgumentParser({
    description:
        'Train an attention-based date-conversion model in TensorFlow.js'
  });
  argParser.addArgument('--gpu', {
    action: 'storeTrue',
    help: 'Use tfjs-node-gpu to train the model. Requires CUDA/CuDNN.'
  });
  argParser.addArgument('--epochs', {
    type: 'int',
    defaultValue: 2,
    help: 'Number of epochs to train the model for'
  });
  argParser.addArgument('--batchSize', {
    type: 'int',
    defaultValue: 128,
    help: 'Batch size to be used during model training'
  });
  argParser.addArgument('--trainSplit ', {
    type: 'float',
    defaultValue: 0.25,
    help: 'Fraction of all possible dates to use for training. Must be ' +
    '> 0 and < 1. Its sum with valSplit must be <1.'
  });
  argParser.addArgument('--valSplit', {
    type: 'float',
    defaultValue: 0.15,
    help: 'Fraction of all possible dates to use for training. Must be ' +
    '> 0 and < 1. Its sum with trainSplit must be <1.'
  });
  argParser.addArgument('--savePath', {
    type: 'string',
    defaultValue: './dist/model',
  });
  argParser.addArgument('--logDir', {
    type: 'string',
    help: 'Optional tensorboard log directory, to which the loss and ' +
    'accuracy will be logged during model training.'
  });
  argParser.addArgument('--logUpdateFreq', {
    type: 'string',
    defaultValue: 'batch',
    optionStrings: ['batch', 'epoch'],
    help: 'Frequency at which the loss and accuracy will be logged to ' +
    'tensorboard.'
  });
  return argParser.parseArgs();
}

async function run() {
  const args = parseArguments();