evaluateModel-tfjscnn.js

function predict(model, data, testDataSize = 500) {
  const testData = data.nextDataBatch(testDataSize, true);
  const testxs = testData.xs.reshape([testDataSize, 28, 28, 1]);
  const labels = testData.labels.argMax([-1]);
  const preds = model.predict(testxs).argMax([-1]);

  testxs.dispose();
  return [preds, labels];
}

async function displayAccuracyPerClass(model, data) {
  const [preds, labels] = predict(model, data);
  const classAccuracy = await tfvis.metrics.perClassAccuracy(labels, preds);
  const container = {name: 'Accuracy', tab: 'Evaluation'};
  tfvis.show.perClassAccuracy(container, classAccuracy, classNames);

  labels.dispose();
}

async function displayConfusionMatrix(model, data) {
  const [preds, labels] = predict(model, data);
  const confusionMatrix = await tfvis.metrics.confusionMatrix(labels, preds);
  const container = {name: 'Confusion Matrix', tab: 'Evaluation'};
  tfvis.render.confusionMatrix(
      container, {values: confusionMatrix}, classNames);

  labels.dispose();
}

async function evaluateModelFunction(model, data)
{
  await displayAccuracyPerClass(model, data);
  await displayConfusionMatrix(model, data);
}