# LSTM data scaling formula based on range

I am planning to run LSTM for a prediction project which has the following input

Presure Temp  DateTime              Strain  Output
123     515   2018-01-11 00:00:10   0.02    1
145     515   2018-01-11 00:00:20   0.03    1
145     515   2018-01-11 00:00:30   0.04    1


My issue is that for the scaling can I base let say for the pressure maximum value is 1000 so I scale based on 123/1000 and I do the same for temp but do I need to scale date time?

• Normally the timestamp of the samples are not directly part of the input to a prediction model - so there would be no gain in doing any scaling. – Shaido Nov 29 '18 at 5:45
• @Shaido thank you for the confirmation. Just wondering if the timestamp is not part of the input then how does it work or help in sequence prediction? Another thing in my case is a simple value/max will be good for my scaling ? – user8012596 Nov 29 '18 at 9:58
• You can take multiple values in a window as input (here the window depends on the time). For example, you could use the data from the latest hour or so. Min-max scaling is implemented in e.g. sklearn: scikit-learn.org/stable/modules/generated/… so you don't need to implement it yourself. Whether it will improve the results depends on the data but it can't hurt to try it out. – Shaido Nov 29 '18 at 10:04
• @Shaido ok I will take your advice on that I saw few different sklearn formula to conduct scaling. I will give it a try on that. What do you mean whether it will improve the results depends on the data but it can't hurt to try it out? – user8012596 Nov 29 '18 at 10:09