is there anything preventing us running GD )as opposed to SGD) on non-average decomposing measures - e.g. F measure?
sorry for the silly question: linear post processing means an afine function of TP, TN? so F measures are not linear?
Thanks. the followup then becomes - how does one find the equivalent measure in practice (since the distribution is not known in advance)
Feel free to unmute and ask if this discussion didn't answer it
thank you! I’ll let it marinate a bit .