this is for holding javascript data
fedhere edited section_Requirements_We_want_the__.tex
almost 8 years ago
Commit id: 00ce7e891ae112beddc971b81b041469395669f4
deletions | additions
diff --git a/section_Requirements_We_want_the__.tex b/section_Requirements_We_want_the__.tex
index 9a2d949..18bbb8b 100644
--- a/section_Requirements_We_want_the__.tex
+++ b/section_Requirements_We_want_the__.tex
...
\item{{\bf Maintain the smoothness requirement, while the uncertainties may be underestimated or overestimated, and fill gaps with simple extrapolations: linear or nearly linear, so as to not to over-interpret the data:} The smoothness requirement can be enforced by minimizing the second derivative of the fit.}
\end{enumerate}
I choose a square exponential kernel ({\tt ExpSquaredKernel} in {\tt george}) $$k(d)=θ_1^2 \sum
{ \left|
{\exp(\frac{−d^2}{θ_2^2})\right|$$. \exp(\frac{−d^2}{θ_2^2})\right| }$$.
In order to choose the hyperparameters $θ_1^2$ and $θ_2^2$ I minimize the objective function: $\chi^2(m) ~*~ \sqrt{\frac{d^2 m}{dt^2} }$, where $m$ is the magnitude, and the fit is done in log time. This works remarkably well in general, with some ``pathological'' exceptions.