Resampling

The DAP is picky about the spectral sampling required to fit the spectra; the spectra must be geometrically sampled in wavelength (constant steps in, e.g., \(\log\lambda\). Ideally, the data-reduction procedures would provide this sampling directly for your data (even if this is a rather unnatural sampling of data from a spectrograph). However, if the spectra are, e.g., sampled linearly in wavelength, the DAP provides a utility to robustly resample the spectra.

Warning

The DAP is, in fact, very picky about the sampling. Even if you expect the sampling is geometric, the DAP may throw an error if the wavelength vector does not yield a constant geometric step to (nearly) numerical accuracy. If that’s the case, please Submit an issue. The likely solution will be to use Resample to resample the data anyway, or you can follow the riskier approach to fool the DAP by remaking the wavelength vector so the step is constant to numerical accuracy (e.g., using numpy.logspace).

The example script test_resample.py illustrates how to use Resample to resample a spectrum. This class is a generalization of the log_rebin function provided by Michele Cappellari in the pPXF package. The script uses the resampling code to de-redshift example spectra from MaNGA, and compares the results to a brute force approach and, if it is installed in your environment, spectres. The script should result in identical results for the brute force, spectres, and Resample; however, Resample is ~100 times faster than the brute-force approach and ~10 times faster than spectres.

For another practical example of how to use Resample, specifically for a full datacube, see the resampling of a KCWI datacube here: Example: KCWI.