tensorflow
Validate the NPV Classification Metric
Recently observed some noise in the NPV metric as the distance increases. Add tests and verify the metric is consistent as the distance increases. The number of predicted negative decreases as the distance increases and this may make the denominator unstable towards the extremes.
Hi, I would like to take up this issue. Please let me know if it's okay to proceed.
@ebursztein Would adding a small epsilon value to stabilize the NPV denominator at extreme distances be the correct approach here, or should we explore other methods?
I am unsure - Owen was the one looking at it. In general the simplest and closest to the paper the better.
@ebursztein Could you please point me to the specific paper you are talking about