search

keflavich / MultiComponentFitting

Sprint repository for multi-component fitting. To be migrated to radio-astro-tools when ready
0 stars 0 forks source link

Merge scousepy #9

Closed vlas-sokolov closed 6 years ago

vlas-sokolov commented 7 years ago

WIP breakdown below:

What does it do?

  1. Averages the input cube in specified spectral averaging areas (SAAs)
  2. Finds the starting guesses for the averaged spectra through two methods: a) multicube: faster parameter space gridding for low to moderate dimensionalities; and b) differential evolution method - I expect it to work better for higher dimensionalities / weird shapes of log-likelihood function.

What it doesn't [yet] do:

  1. It doesn't yet wrap the spectral guesses back into the original cube,
  2. doesn't yet fit them with pyspeckit fitting routines,
  3. No model selection switchers are in place yet.
  4. Interactive guess verification and tinkering from the user side.

1 and 2 should be very easy to add, 3-4 should take some effort to write in. So far it's possible to run the two example functions to get the feel of how the two global guess setters (to be fair DE is more like a fitter on its own) behave.

keflavich commented 7 years ago

@vlas-sokolov what version of multicube does this depend on? I got several breakages trying to run the examples. Does it require the guided-exceptions-fiteach branch?

vlas-sokolov commented 7 years ago

@keflavich it might be that not all the examples are up to date... And I'm not around to run any tests to see, nor do I have a decent internet connection. Have you run it with the redefined-fiteach branch? That's the one that's up to date. I've set it to be the main branch on github, but it's not master yet because the idea of the fiteach method being overwritten is pretty bad (it has some experimental features that I'm using once in a while) .

keflavich commented 7 years ago

@vlas-sokolov I was using the wrong branch. I'll try again.