Open adematti opened 2 years ago
There are currently different implementations for constructing delta field ~ data / randoms - 1 (smoothing of the data and randoms fields before computing delta, or after computing delta, threshold in number of randoms per cell), which follow the prescriptions used in original Bautista and White codes, e.g.: https://github.com/cosmodesi/pyrecon/blob/95498903c622bbb38bbd923a56a7f18b5d40d6a0/pyrecon/iterative_fft_particle.py#L50 vs. https://github.com/cosmodesi/pyrecon/blob/95498903c622bbb38bbd923a56a7f18b5d40d6a0/pyrecon/recon.py#L192 It'd be good to converge on a common prescription. Gaussian mocks (e.g. https://github.com/cosmodesi/pyrecon/blob/main/nb/e2e_examples.ipynb) with selection function in them may assist the choice. WIthin current implementation, if randoms density is too small there will be a lot of delta points set to 0, hence highly inefficient reconstruction.
There are currently different implementations for constructing delta field ~ data / randoms - 1 (smoothing of the data and randoms fields before computing delta, or after computing delta, threshold in number of randoms per cell), which follow the prescriptions used in original Bautista and White codes, e.g.: https://github.com/cosmodesi/pyrecon/blob/95498903c622bbb38bbd923a56a7f18b5d40d6a0/pyrecon/iterative_fft_particle.py#L50 vs. https://github.com/cosmodesi/pyrecon/blob/95498903c622bbb38bbd923a56a7f18b5d40d6a0/pyrecon/recon.py#L192 It'd be good to converge on a common prescription. Gaussian mocks (e.g. https://github.com/cosmodesi/pyrecon/blob/main/nb/e2e_examples.ipynb) with selection function in them may assist the choice. WIthin current implementation, if randoms density is too small there will be a lot of delta points set to 0, hence highly inefficient reconstruction.