Explain error histogram ?

Hi

I'm using neural network matlab tool

it's hard to understand error histogram with 20bins

what is bar means?where is the best performance? please tell me why

and what is means zero error?

I attatched my graph. please reply ASAP.

Thank you.

Answer 1

[Adam Danz]

I don't know how this plot was created but judging by the axis labels I can estimate what the storyline is. Below is a very general description that you can interpret under your NN's context.

It looks like something was tested many times (my estimate is ~700 iterations) and on each time the error was computed between Targes and Outputs (from the xlabel). All of the error was quite low and falls in the range of -1.1e07 to 3.29e08 but most of the error falls within the range of x-values where you see the high bars. The height of each bar is the number of tests (ie, 'instances').

Notice that the orange line "zero error" separates negative and positive ticks. The sign of the error shows you the direction of the bias. Positive error means the outputs were less than the target and negative error means that the outputs were greater than the target (assuming the targets are positive values) . I base this on the x-axis label that defines error as Targets-Outputs.

The majority of tests were negative so the marjority of outputs were slightly larger than the targets.

[ayodeji ogundana]

quoting your response

"Positive error means the outputs were smaller than the targets and negative error means that the targets were larger than the outputs (assuming the targets are positive values)".

I think you just repeated same thing in different way?

[Adam Danz]

Thanks @ayodeji ogundana, I fixed that in my answer. Good eye!

Answer 2

[Abhilash Singh]

In addition to the above explanation, the readers may also visit the following article to know how to interpret the error histogram (https://www.mdpi.com/1424-8220/22/3/1070).