Please help understand how accurate or reliable
tf.keras.metrics.AUC metric is.
tf.keras.metrics.AUC is not actually calculating the true AUC but approximating it.
This classes approximates AUCs using a Riemann sum: During the metric accumulation phrase, predictions are accumulated within predefined buckets by value. The AUC is then computed by interpolating per-bucket averages. These buckets define the evaluated operational points.
There is a comment that such calculation is not valid.
Using AUC as metric doesn't work because Keras calculates the AUC for each minibatch and average the results, such calculation is not valid for AUC