Utils for binning of spectra. #52
Conversation
The complex stellar population subpackage is meant to combine bpass variables with a stelalr formation history to generate more complex populations. To this end, we're implementing a stellar formation history class. The basic SFH class for custom SFH is implemented using scipy splines. This basic class is also tested.
This method allows for the calculation of the event rate at a given lookback time. It uses the BPASS binning for this. This method is much faster, but suffers from a lack of accuracy.
Both functions are implemented for the BPASS spectra. However, the method is similar to the BPASS event rate calculation, which is sub optimal for the spectra calculation. It takes an extremely long time to do the calculation. A for loop of 100000 iterations is present, because each wavelength is seen as it's own "event type". It should be possible to remove this by implementing a function to take the separate wavelengths as a single unit. It would be interesting so see the influence of putting the for loop within a numba function, but thinking about restructuring the function is probably better. Including a "caching" of the BPASS spectra into a pickled DataFrame. Additional test files are required for this to run.
Instead of requiring a scipy interpolated spline as input. These functions now take a list of functions as input for the stellar formation and metallicity history. This adjustment has been made by using `numpy.interp` instead of scipy.interpolate. Furthermore, instead of scipy's spline integration `numpy.trapz` is now used to calculate the mass per bin.
* BUG: use np instead of pd in SpectraCompiler. * BUG: fix index and make test pass. * BUG: fix another regression bug. * Fix further regression bug. Authored-by: Martin Glatzle <mglatzle@mpa-garching.mpg.de>
Also includes some docstring updates
CSPSpectra now has the following functions: Public: - at_time: at time with function input - over_time: over lb time with function input - grid_at_time: at time with SFH grid input - grid_over_time: over lb time with SFH grid input Private: - grid_rate_calculator_at_time - grid_rate_calculator_over_time The grid_rate_calculators_* are numba functions that have the grid_* around them as wrappers for the input and output
Instead of N bins being created, N-1 bin was being created.
There was a problem hiding this comment.
Excellent work and must apologise for the time it took me to get to this: I was taking a course on deep learning and then had to take sick leave to treat my HVS.
I've added a few comments to the files and I have one main questions: is there a reason you added these function to the utils module and not the spec module which specifically has to do with spectroscopy?
|
No worries, thanks for taking a look! Regarding the location in Will address the other issues as time permits. |
renamed variables: - BPASS_METALLICITIES. now with `z` renamed functions: - all_rates -> rates_all_z - all_spectra -> spectra_all_z
|
Merged |
great, that makes sense :) |
Re-implementation of the binning utilities of pyBASS using numba. I haven't quite managed to achieve the same performance as that of the original Fortran implementation, but this should not be a problem even if binning millions of spectra.