Following some of our preliminary experiments listed in #2 I think we have a few good ideas and threads to explore, but we should start thinking about concrete milestones and goals in order to be where we want to be by the end of January. I'm opening this issue to try to track our various todos.
Goal
By end of January we want to have plots comparing the dependence on number of sampling points of various density estimators to achieve a given accuracy in the modeling of likelihood or posterior distributions. We can push this to various levels:
Level 0: Basic requirements: On a toy model like two moons, or fake weak lensing likelihood toy model, make a plot of number of sims vs accuracy of modeling with existing methods (e.g. various kind of flows).
Level 1: Main goal: Develop a density estimator that can make use of gradient information, add comparison plot to Level 0 plots
Level 2: Stretch goal: Apply the developed density estimator to a realistic WL simulation problem.
TODOs
Method developments:
[x] #6
[x] #7
[ ] #13
[ ] #20
Simulation developments:
[x] #8
[ ] #19
[ ] #18
Comparison infrastructure:
[ ] #17
[ ] Add scripts to make plots comparing different NDEs on different distributions
Following some of our preliminary experiments listed in #2 I think we have a few good ideas and threads to explore, but we should start thinking about concrete milestones and goals in order to be where we want to be by the end of January. I'm opening this issue to try to track our various todos.
Goal
By end of January we want to have plots comparing the dependence on number of sampling points of various density estimators to achieve a given accuracy in the modeling of likelihood or posterior distributions. We can push this to various levels:
Level 0: Basic requirements: On a toy model like two moons, or fake weak lensing likelihood toy model, make a plot of number of sims vs accuracy of modeling with existing methods (e.g. various kind of flows).
Level 1: Main goal: Develop a density estimator that can make use of gradient information, add comparison plot to Level 0 plots
Level 2: Stretch goal: Apply the developed density estimator to a realistic WL simulation problem.
TODOs
Method developments:
Simulation developments:
Comparison infrastructure: