tensorflow自定义梯度

发表于 | 阅读次数: 
1
2
3
4
5
6
7
opt = GradientDescentOptimizer(learning_rate=0.1)

grads_and_varas = opt.compute_gradients(loss, list_of_vars)

capped_grads_and_vars = [(fun(gv[0]), gv[1]) for gv in grads_and_vars]

opt.apply_gradients(capped_grads_and_vars)
请作者喝一杯咖啡☕️