多分类focal loss

Focal Loss的介绍见 Focal Loss(RetinaNet)原理与代码解析_00000cj的博客-CSDN博客

from sklearn.preprocessing import LabelBinarizer
from math import log

import tensorflow as tf
import os
os.environ['CUDA_VISIBLE_DEVICES'] = str(0)
tf.enable_eager_execution()

labels = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']  # 所有标签
y_true = ['1', '4', '5']  # 样本的真实标签
y_pred = [[0.1, 0.6, 0.3, 0, 0, 0, 0, 0, 0, 0],
          [0, 0.3, 0.2, 0, 0.5, 0, 0, 0, 0, 0],
          [0.6, 0.3, 0, 0, 0, 0.1, 0, 0, 0, 0]]  # 样本的预测概率

lb = LabelBinarizer()
lb.fit(labels)
transformed_labels = lb.transform(y_true)
# print(transformed_labels)
# [[0 1 0 0 0 0 0 0 0 0]
#  [0 0 0 0 1 0 0 0 0 0]
#  [0 0 0 0 0 1 0 0 0 0]]

""" multiclass cross-entropy loss """
N = len(y_true)  # 样本个数
K = len(labels)  # 标签个数

eps = 1e-15      # 预测概率的控制值
Loss = 0         # 损失值初始化

for i in range(N):
    for k in range(K):
        # 控制预测概率在[eps, 1-eps]内，避免求对数时出现问题
        if y_pred[i][k] < eps:
            y_pred[i][k] = eps
        if y_pred[i][k] > 1-eps:
            y_pred[i][k] = 1-eps
        # 多分类问题的交叉熵计算公式
        Loss -= transformed_labels[i][k] * log(y_pred[i][k])

Loss /= N  # 交叉熵loss


""" multiclass focal loss"""
epsilon = 1e-9
y_true = transformed_labels
gamma = 2
alpha = 4
# alpha = [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1]

y_true = tf.convert_to_tensor(y_true, tf.float32)
y_pred = tf.convert_to_tensor(y_pred, tf.float32)

model_out = tf.add(y_pred, epsilon)
ce = tf.multiply(y_true, -tf.log(model_out))

weight = tf.multiply(y_true, tf.pow(tf.subtract(1., model_out), gamma))

fl = tf.multiply(alpha, tf.multiply(weight, ce))
reduced_fl = tf.reduce_max(fl, axis=1)
focal_loss = tf.reduce_mean(reduced_fl)

参考

多分类问题的交叉熵计算 - SegmentFault 思否