Closed erikbern closed 6 years ago
this seems to improve convergence for some real world data i have
not sure why/if it's so much slower, should figure out if piping it through feed_dict is the wrong way to do minibatches
feed_dict
Coverage remained the same at 94.35% when pulling 87a900a975eb16024842b2ac6fa155edec9aba00 on lower-batch-size into e5eec9ea1bb292e1e9a8dbd6ff7e17ded5ef4a5b on master.
this seems to improve convergence for some real world data i have
not sure why/if it's so much slower, should figure out if piping it through
feed_dict
is the wrong way to do minibatches