on 07-10-2017 10:11 AM
Hello,
I want to ask some question about SAP Predictive Analytics time series.
1.What is the difference between “Linear in Time” and “Autoregressive” ?
2.Autoregressive use previous target to predict future target, but I don’t know what is “Linear in Time” meaning?
3.What is the “Weight” meaning in Selecting Variables,is there any example can help me understand this function? and what “Weight Quantum” is in General model settings?
4.I try to change the value "Percentage of Variable Contributions to Keep", when I enter 100,it means that trends will keep all extra-predictable variable, is that correct? and I try to decrease the value as 30 or lower one, but it still keep all extra-predictable variable in the result, so I confused, I think it should keep fewer extra-predictable variable at trend model, do you know what happend?
if anyone knows, please don't hesitate to help me.
thanks a lot.
1. Linear in Time means it fits a regression line to the target data, using the Time variable as the X variable.
2. Auto regressive means it performs a linear combination on the de-trended data, using past values of the target variable itself.
3. There is no use case to my knowledge for a weight in time series
4. With default value 95%, I get variables with a least 5% contribution ; 98%, with at least 2% contribution ; for some reason if I go below the default value (that is 95%) I get all the tiny contributions as if I entered 100%
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
hello Marc, thanks for your answer.
In my fourth question, I change the value "Percentage of Variable Contributions to Keep" to 60, so that all variable used in trend model's contribution should at least 40%, but still see the variable contributions at "Display > Regressions:Contributions by Variables" under 40%, I understand aright?
User | Count |
---|---|
85 | |
10 | |
9 | |
8 | |
6 | |
6 | |
6 | |
5 | |
3 | |
3 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.