This is a small server collapse for a while. . .
Completed today
Because yesterday had already seen a tendency fitting, so today without regularization first part, first find the best model
Still two hidden layer, adjusting node parameters n
This is 64X64
This is 64X32
This is 32X32
This is 16X16
Look out 16X16 like a lot of potential, or being not too fit?
So the first training
Tomorrow Plan
This model of tomorrow should adjust the parameter, then add regularization to prevent over-fitting, have to do something within this week, was set aside one week to write papers.
Impressions today
Sure enough, what method is most important, there are only good way, according to this model a little bit to do on the line