Bayes by Backprop with gluon jupyter notebook


#1

Hi,

Going over the notebook here and it seems that the weights custom update step in the training loop doesn’t work (errors out because param._data.keys() method doesn’t exist).

Is this a known issue? Worth investigating and trying to fix?

Thanks,
Yevgeniy


#2

Yeah that code is wrong. Accessing private members (_data) is bad practice. Should use param.data(ctx) instead.