search

mattjshannon / auto-spline

Automatic generation of a spline continuum for astronomical spectra.
1 stars 0 forks source link

Implementing tensorflow (keras?) method #4

Open mattjshannon opened 6 years ago

mattjshannon commented 6 years ago

Issue -- deciding which type of method to use for this problem. Mostly using this comment thread to keep track of my progress.

mattjshannon commented 6 years ago

From the astroinformatics school, Xiangru Li suggested a FFN (feed-forward neural network) method. Also pointed out that a convolutional neural network (CNN) is likely a bad choice for this sort of problem.

mattjshannon commented 6 years ago

Lee Giles suggested logistic regression and/or random forest would be a good approach for this problem.

Tom Loredo mentioned the possibility of L1 Lasso and then using an MCMC. Possibly a shotgun stochastic search (possibly reversible jump MCMC?).