Hi Nayoung,
Typically errors are calculated with L1 (abs(errors)) or L2 ((errors)^2) norms. Meaning that it cancels out signs from difference between target and output. The above bar graph doesn't have neither absolute or square for errors, it has negative errors which isn't typical. Nonetherless errors are close to zero meaning the output is close to target value, the NN models well.
- The height of bar in the Bar plot means how many data points are near the bin value.
- 0.001502 is selected error value based on bin number setup in histogram function (i.e. histogram(data,20)), it can be changed if you use different bin value. You can try bin number with 50 or 100.
- 0.001502 is the closest number to 0 among those bin numbers. Because error = 0 means the ourputs are close to the target, so model performance is good if the bin bars close to zero are taller than higher error numbers.
Hope this helps you.
Best,
IK
