FP, FN and TP for a multiclass classification. Do we need negative image ( images not in dataset) before we can get TN? Please relevant reference would be appreciated.. Thanks in anticipation.
Hi. One of the easiest step could be just to calculate metrics for each class separately and average them for an aggregation usages, if needed
Thanks for the response.... So no need for negative image?
well, it depends on what are you doing. But it could be a nice feature as well
When it comes to multiclass classification, the True Negative images are the images that do not belong to the class you are trying to assess. I.e. all other images are TN images. Consider you have N classes [C_1, C_2, ... C_n]. So in order to compute TN rate for the C_1, you treat images that belong to C_2, C_3, ... C_N as TN images. Therefore you do not need to have a separate set of TN images.
Ive talken with him a bit and it seems like he meant that he need to fit some task outline cases for getting zeros for all classes or smth like that
Обсуждают сегодня