An example would be predicting hours on a 24 hour clock. 1 and 24 are only 1 hr apart. Is there any way to adjust for this when training so the nn doesn't think it is off by 23 hrs?
Why don't you just (for example) convert it to binary or so?
But I honestly don't get your question, so maybe my suggestion is crap…
This is actually a preprocessing problem and not a NN problem.
(Well, it CAN be but this is what I can suggest from the info the OP gives to us)
If now is 1h, 24h was 1 hour past.
if now is 24h, 1h was 23 hours past.
Do some preprocessing to your data to account for this effect.
Log in to post a comment.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.