mxnet.gluon中trainer.step()的用法

step(batch_size, ignore_stale_grad=False) method of mxnet.gluon.trainer.Trainer instance
    Makes one step of parameter update. Should be called after
    `autograd.compute_gradient` and outside of `record()` scope.
    
    Parameters
    ----------
    batch_size : int
        Batch size of data processed. Gradient will be normalized by `1/batch_size`.
        Set this to 1 if you normalized loss manually with `loss = mean(loss)`.
    ignore_stale_grad : bool, optional, default=False
        If true, ignores Parameters with stale gradient (gradient that has not

        been updated by `backward` after last step) and skip update.

 上面是help打出来的文档，可见到有两个参数，第一个参数是batch_size，

第二个如果是true的话，就忽略那些参数，哪些参数呢，即它的梯度还没有被后向传播更新，也即stale gradient

，并且忽略更新，

默认情况下是false。