Function drawROC

website-phishing/index.js:64–90

* Draw a ROC curve. * * @param {tf.Tensor} targets The actual target labels, as a 1D Tensor * object consisting of only 0 and 1 values. * @param {tf.Tensor} probs The probabilities output by a model, as a 1D * Tensor of the same shape as `targets`. It is assumed that the values of * the

(targets, probs, epoch)

 * @returns {number} Area under the curve (AUC).
 */
function drawROC(targets, probs, epoch) {
  return tf.tidy(() => {
    const thresholds = [
      0.0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4,  0.45, 0.5,  0.55,
      0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.92, 0.94, 0.96, 0.98, 1.0
    ];
    const tprs = [];  // True positive rates.
    const fprs = [];  // False positive rates.
    let area = 0;
    for (let i = 0; i < thresholds.length; ++i) {
      const threshold = thresholds[i];

      const threshPredictions = utils.binarize(probs, threshold).as1D();
      const fpr = falsePositiveRate(targets, threshPredictions).dataSync()[0];
      const tpr = tf.metrics.recall(targets, threshPredictions).dataSync()[0];
      fprs.push(fpr);
      tprs.push(tpr);

      // Accumulate to area for AUC calculation.
      if (i > 0) {
        area += (tprs[i] + tprs[i - 1]) * (fprs[i - 1] - fprs[i]) / 2;
      }
    }
    ui.plotROC(fprs, tprs, epoch);
    return area;
  });
}

// Some hyperparameters for model training.
const epochs = 400;