Focal_loss 实现

 def focal_loss(logits,onehot_labels,  gamma=2.0, alpha=4.0):

    
     """
    
     focal loss for multi-classification
    
     FL(p_t)=-alpha(1-p_t)^{gamma}ln(p_t)
    
     Notice: logits is probability after softmax
    
     gradient is d(Fl)/d(p_t) not d(Fl)/d(x) as described in paper
    
     d(Fl)/d(p_t) * [p_t(1-p_t)] = d(Fl)/d(x)
    
     Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017).
    
     Focal Loss for Dense Object Detection, 130(4), 485–491.
    
     https://doi.org/10.1016/j.ajodo.2005.02.022
    
     :param labels: g