Repository navigation
added bayesian information criterion and akaike information criterion - #4716
Conversation
|
Some feedback from @jakevdp might also be helpful here? |
|
Looks good – a couple minor comments:
|
|
Hi @jakevdp, thanks! What does "standard version" stand for? (Root) Mean Square Error? |
I just meant that you could call the |
|
@jakevdp Sure, understood now, thanks =) |
|
Are these failures unrelated with this PR, perhaps? |
|
@mirca - Could you please rebase? It seems that there are some conflicts, thus the ci doesn't even start. |
|
(If you haven't done a rebase before and need help, I can assist with in via gitter chat in the next hour or so, or tomorrow.) |
|
@bsipocz Thanks =D I did a couple rebases before, I think this seems to be fine too. |
|
👍 |
|
Looks good! |
|
|
||
| in which :math:`n` is the sample size, :math:`k` is the number of free | ||
| parameters and :math:`\mathrm{SSR}` stands for the sum of squared redisuals | ||
| between model and data. |
There was a problem hiding this comment.
can you give a reference for this?
|
@crawfordsm I think it's all fine here now :) |
|
LGTM |
5 participants
Hi all,
This is an attempt to finish up the work started by @astronomeralex at #3700
Basically, these information criterions are used to decided whether increasing the number of parameters in a model truly improves the fitting. Conversely, they are also used to verify whether improvements in fitting are due to the increasing of the number of parameters.
Nice readings about these criterions in an astrophysical context:
In the examples, I tried to show how these information criterions might be used together with astropy fitters.
Appreciate any thoughts.
@cdeil @crawfordsm @astrofrog @hamogu @embray (pinging the ones involved on the old PR).