Open kisstea opened 9 years ago
Yes, your are right, The paper didn't mention this either. I advice you set the recall with 1 in some stages, If have any good strategies. please write down here
I think you can try the soft cascade method to set recall
I find the code removes positive samples after each CART as the paper writes. My question is that, even I set a high precision-recall such as 0.999. After 5 stages, the positive remained is 0.999^(5040)=0.006457. I wonder if the detection rate is too poor?