sato-cloudian · December 17, 2015 06:10
diff --git a/HOW_TO_READ_DL4J_UI b/HOW_TO_READ_DL4J_UI
 Scores vs. iteration
 1. See if "Scores vs. iteration" decends quickly, and gets flat as iterations increase
 2. If it becomes flat too early like above 1.0, that means it would have some rooms of tuning.

 Model/Gradient
 1. See if they show a normal distribution because weights&biases are converged on certain values
 2. Note that a distributions may not look good if you have few samples like Iris

 Mean Magnitudes: Parameters and Updates
 1. Parameter Mean Magnitudes shows an ascend, and gets flat as those values are converges on certain values
 2. Update/Gradient Mean Magnitudes shows a decend, and get flat as those values are converges on certain values
	Scores vs. iteration
	1. See if "Scores vs. iteration" decends quickly, and gets flat as iterations increase
	2. If it becomes flat too early like above 1.0, that means it would have some rooms of tuning.

	Model/Gradient
	1. See if they show a normal distribution because weights&biases are converged on certain values
	2. Note that a distributions may not look good if you have few samples like Iris

	Mean Magnitudes: Parameters and Updates
	1. Parameter Mean Magnitudes shows an ascend, and gets flat as those values are converges on certain values
	2. Update/Gradient Mean Magnitudes shows a decend, and get flat as those values are converges on certain values
No results found