can I use this project in time series prediction ?

[dswah]

can I use this project in time series prediction ?

[dswah]

@mathcbc
yes, of course, but it will only be capable of AR (auto-regressive) models, which is a reasonable place to start.

[dswah]

@mathcbc however, the package is still very immature. this is why i have no installation docs.
for example extrapolation is currently broken.

i would ask you to wait a few days before using the code.

[mathcbc]

AR models are linear so they are not what i want. Additive models have smooth functions of variable X_i, so they can solve nonlinear models. Do your algorithm contains smoother？ At 2017-03-11 02:01:49, "daniel servén" <[email protected]> wrote: @mathcbc yes, of course, but it will only be capable of AR (auto-regressive) models, which is a reasonable place to start. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

[dswah]

Yes, the algorithm learns a smoother on each feature.

What i mean is that you may want to choose the number of AR lags a-priori and structure your feature vector so that each lag is a feature. This way the GAM will learn a smoother for each lag.

In effect this is a sort of non-linear AR model:

x_t = f_1(x_{t-1}) + f_2(x_{t-2}) + ... + f_p(x_{t-p}) + c

just a thought. :)