* a Gaussian Process is a drop
-
in
replacement for chi squared
* model building & computational
cost are (solvable!) challenges
* you should check out tinygp!
Slide 31
Slide 31 text
case study: probabilistic inference
2
Slide 32
Slide 32 text
have:
physics
= >
data
Slide 33
Slide 33 text
want:
data
= >
physics
Slide 34
Slide 34 text
7
[1] physical models
[2] legacy code
Slide 35
Slide 35 text
No content
Slide 36
Slide 36 text
number of parameters
patience required
a few tenish not outrageously many
reference: DFM (priv. comm.)
Slide 37
Slide 37 text
number of parameters
patience required
emcee
a few tenish not outrageously many
reference: DFM (priv. comm.)
Slide 38
Slide 38 text
number of parameters
patience required
emcee
a few tenish not outrageously many
how things should be
reference: DFM (priv. comm.)
reference: Soichiro Hattori, Ruth Angus, DFM,
. . .
(in prep)
showing 23 of the
404 parameters
(8 per channel + 4 shared)
Slide 45
Slide 45 text
how?
Slide 46
Slide 46 text
d(physics
= >
data) / dphysics
Slide 47
Slide 47 text
automatic differentiation
aka “backpropagation”
Slide 48
Slide 48 text
No content
Slide 49
Slide 49 text
7
[1] physical models
[2] legacy code
Slide 50
Slide 50 text
7
[1] domain
-
specif
i
c libraries
[2] emulation
Slide 51
Slide 51 text
No content
Slide 52
Slide 52 text
* gradient
-
based inference using
autodiff can improve eff
i
ciency
* there are practical challenges
with these methods in astro
* of interest: domain
-
specif
i
c
libraries & emulation
Slide 53
Slide 53 text
aside: JAX
3
Slide 54
Slide 54 text
No content
Slide 55
Slide 55 text
import numpy as np
def linear_least_squares(x, y)
:
A = np.vander(x, 2)
return np.linalg.lstsq(A, y)[0]
Slide 56
Slide 56 text
import jax.numpy as jnp
def linear_least_squares(x, y)
:
A = jnp.vander(x, 2)
return jnp.linalg.lstsq(A, y)[0]
Slide 57
Slide 57 text
No content
Slide 58
Slide 58 text
open research practices
4
Slide 59
Slide 59 text
No content
Slide 60
Slide 60 text
No content
Slide 61
Slide 61 text
No content
Slide 62
Slide 62 text
No content
Slide 63
Slide 63 text
No content
Slide 64
Slide 64 text
No content
Slide 65
Slide 65 text
No content
Slide 66
Slide 66 text
open software is foundational to
astrophysics research
there are opportunities at the
interface of astro & applied f
i
elds
there are ways you can participate
& benef
i
t right away
Slide 67
Slide 67 text
7
I want to chat about…
[1] your data analysis problems
[2] building astronomical software
[3] writing documentation & tutorials