Working feed_dict use in conjunction with the Dataset object. This seems decently fast, but either way, I don't know how to avoid it, considering that calling the ops which use the Dataset batches consumes the value and they can't be used again.
I have also implemented a way to easily pipe in the MNIST data to evaluate the model. I will work on adding more "overriders" to make the MNIST reliably converge.
Working
feed_dictuse in conjunction with theDatasetobject. This seems decently fast, but either way, I don't know how to avoid it, considering that calling the ops which use theDatasetbatches consumes the value and they can't be used again.I have also implemented a way to easily pipe in the MNIST data to evaluate the model. I will work on adding more "overriders" to make the MNIST reliably converge.
Added a histogram summary for input
xdata.