Lambda values
jrgant opened this issue · comments
Each run of the quadratic trend filter produces a set of lambda values. Currently, I'm using one randomly selected lambda value for each of the 14 seasons, based on the fitted trend filter.
However, I'm not certain what this lambda value represents, just that it's needed to produce the predicted curves and that different lambdas produce different predictions.
If we need to account for all lambdas in order to propagate uncertainty into the curve distributions, I'll need to update the code to loop through these values as well.
References of potential relevance:
Tibshirani RJ. Adaptive piecewise polynomial estimation via trend filtering. Ann Stat 2014;42:285–323. doi:10.1214/13-AOS1189.
Ramdas A, Tibshirani RJ. Fast and Flexible ADMM Algorithms for Trend Filtering. arXiv [statML] 2014.
I think I understand the basic issue, not sure the best approach though. Could you direct me to the code section? I think this is line 162 in curvefit.R?