major restructuring of parameter distribution and design of the inter… #38
Conversation
…polation grid for the kinematic scaling
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #38 +/- ##
==========================================
- Coverage 99.40% 99.37% -0.03%
==========================================
Files 46 48 +2
Lines 2346 2407 +61
==========================================
+ Hits 2332 2392 +60
- Misses 14 15 +1
|
…kin_scaling_restructuring # Conflicts: # test/test_Sampling/test_Distributions/test_lens_distribution.py
| # we draw a linear gaussian for 'const' anisotropy and a scaled proportional one for 'OM | ||
| if self._distribution_function in ["GAUSSIAN"]: | ||
| if self._anisotropy_model == "OM": | ||
| a_ani_draw = np.random.normal(a_ani, a_ani_sigma * a_ani) |
There was a problem hiding this comment.
For my own understanding, what's the reasoning for a scaled proportional Gaussian for OM? Also, would it be more clear/explicit if you introduce another distribution type (e.g., "SCALED GAUSSIAN")?
There was a problem hiding this comment.
the main reason was as a_ani close to zero, the distribution becomes one-sided and has a much stronger effect, so having a relative scatter distribution seemed more reasonable (and recommended). I understand that this was not transparent in the code and I changed it now to GAUSSIAN_SCALED. I suggest you use this when using "OM" or "GOM" models
…i with scaled function separately
…kin_scaling_restructuring
…kin_scaling_restructuring
…kin_scaling_restructuring # Conflicts: # test/test_LensPosterior/test_kin_scaling_config.py
…kin_scaling_restructuring
…polation grid for the kinematic scaling
New: require list of parameter names with the interpolation grid, which allows more flexibility for further development