Open atrophiedbrain opened 5 years ago
The default Epanechnikov_kernel returns 0 if abs(t) > 0. All my time points are positive. When I replaced my time vector with const times = collect(7.0:time_step:85.0)/85.0 the two_stage_method function runs sorry it returns 0 if abs(t) > 1 so this line in two_stage_method was returning 0 for all of my time points: W[i] = kernel_function((tpoints[i]-t)/h)/h h = (n^(-1/5))(n^(-3/35))((log(n))^(-1/16)) where n is the number of time points this is a decaying function that is less than 1, so (tpoints[i]-t)/h will be greater than 1 for all of my times that are greater than one Switching to a kernel that doesn't return 0 for time values > 1 works, for example Logistic_Kernel.
@ChrisRackauckas commented that perhaps we should change the default kernel so that the default kernel supports time series > 1.
I am receiving a LAPACKException when using two_stage_method. A MVE is included in this post.
Exception:
Both calls to two_stage_method in the MVE below, one using parameters and one not, both return this LAPACKException.