support vector regression practice
2018.07.16 update
Test SVR regressioin on time series energy dataset on UCI
Time step = 5, adjust to 7
Use 7 days of energy and weather data to predict 8th day's energy consume in a low energy house.
Result:
Final argument: C = 100, time step = 7
Discussion:
We could set penalty C be large to 100, and SVR could still get a proper result on test set. First use linear kernel, and found performance is poor, then adjust to RBF kernel.
SVR is availabel for time series forecasting.