Closed j-faria closed 3 years ago
Standard Gaussian priors should work most times, but not always.
Maybe go back to Data.topslope()
?
Default priors are Gaussian with mean 0 and sigma 10^m, where m is the order of magnitude of the coefficient, as supported by the data:
(int) round( log10( get_RV_span() / pow(get_timespan(), degree) ) );
About the default linear trend option. The behaviour now is
with
const bool trend = true;
const int degree = 0;
the code throws a std::logic_error
with message
trend=true but degree=0, what gives?
but note that
const bool trend = false;
const int degree = 1;
will compile and run, without the trend.
When degree
is larger than 3, there is another std::logic_error
saying
can't go higher than 3rd degree trends
Allow for quadratic (and maybe cubic) trends