def optim(self, learning_rate):
w, b = self.w, self.b
# print("fc layer, optim learning_rate", learning_rate)
w.data, w.m, w.v, w.t = adam(w.data,
w.grad,
learning_rate=learning_rate,
m=w.m,
v=w.v,
t=w.t)
b.data, b.m, b.v, b.t = adam(b.data,
b.grad,
learning_rate=learning_rate,
m=b.m,
v=b.v,
t=b.t)
def sgd_fun_adam_avg(obj_fun, grad_fun, x0, max_iters, callback, lr_fun):
return opt.adam(obj_fun,
grad_fun,
x0,
max_iters,
callback,
lr_fun,
avgdecay=0.99)
def sgd_fun_adam_avg(obj_fun, grad_fun, x0, max_iters, callback, lr_fun):
return opt.adam(obj_fun, grad_fun, x0, max_iters, callback,
lr_fun, avgdecay=0.99)