Add Deviate class that draws from an arbitrary P(x) distribution.

[rmjarvis]

Useful for when we have an arbitrary (e.g. realistic) P(e) function for the ellipticity.

[msimet]

I have C++ code for this--could adapt to or wrap in Python.

[rmandelb]

@msimet Melanie, I wanted to point out two relatively new things in GalSim that should be useful for this:

• A while ago, but possibly after this issue was opened, Mike made his 1d integrator accessible in python (integ1d; for an example of its use see galsim/lensing.py).
• On the branch for Allow PowerSpectrum to accept tabular P(k) #305, Mike made it possible to interpolate a tabulated 1d function.

So I think the code for this issue #306 on drawing random numbers from an arbitrary distribution could be implemented in python by:
(1) reading in your x and P(x) as numpy arrays.
(2) Integrating P(x) with integ1d to get the cumulative probability distribution P(<x).
(3) Make a table out of x vs. P(<x)
(4) sample your uniform random numbers using our standard RNG procedures.
(5) interpolate on the table to get the value of x corresponding to that P(<x).

In practice I think that you might want to make a class called probdist (or something) that is initialized from numpy arrays and does steps (1)-(3), then it can have a method that draws random numbers by doing (4) and (5).

(For context, for anyone else reading this, Melanie mentioned that her code that does this is not in C/C++ after all, so she can't just adapt and wrap it. That's why I'm suggesting how to simply do this from scratch using these new tools in GalSim.)

[msimet]

Question for the group at large: what should we call this? I was thinking about UserDeviate, but knowing how people tend to abbreviate things I worry about confusion with UniformDeviate.

[rmandelb]

I don't have a strong opinion. I agree there's potential for confusion with people referring to 'ud' but this is not a very big deal. Another option might be DistDeviate (i.e., deviate from a distribution).

On Jan 15, 2013, at 1:48 PM, msimet notifications@github.com wrote:

Question for the group at large: what should we call this? I was thinking about UserDeviate, but knowing how people tend to abbreviate things I worry about confusion with UniformDeviate.

—
Reply to this email directly or view it on GitHub.

---

Rachel Mandelbaum
rmandelb@andrew.cmu.edu
http://www.andrew.cmu.edu/~rmandelb/

[barnabytprowe]

Either seems pretty good to me. Leave your mark! :)

On 16 Jan 2013, at 03:20, rmandelb wrote:

I don't have a strong opinion. I agree there's potential for
confusion with people referring to 'ud' but this is not a very big
deal. Another option might be DistDeviate (i.e., deviate from a
distribution).

On Jan 15, 2013, at 1:48 PM, msimet notifications@github.com wrote:

Question for the group at large: what should we call this? I was
thinking about UserDeviate, but knowing how people tend to
abbreviate things I worry about confusion with UniformDeviate.

—
Reply to this email directly or view it on GitHub.

---

Rachel Mandelbaum
rmandelb@andrew.cmu.edu
http://www.andrew.cmu.edu/~rmandelb/
—
Reply to this email directly or view it on GitHub.