Closed asalvadorsalas closed 3 years ago
The use of 4 hidden layers 196 nodes 40% dropout 2048 batch size learning rate 0.00428095 returns a dobule peaked NN score Check if its for different regions, decay or what From different peaks maybe can be reopened
in single GPU every epoch takes < 10 s
The use of 4 hidden layers 196 nodes 40% dropout 2048 batch size learning rate 0.00428095 returns a dobule peaked NN score Check if its for different regions, decay or what From different peaks maybe can be reopened
in single GPU every epoch takes < 10 s