New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
docs: learn nb on covariance matrices #1233
base: main
Are you sure you want to change the base?
Conversation
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Codecov Report
@@ Coverage Diff @@
## master #1233 +/- ##
=======================================
Coverage 97.45% 97.45%
=======================================
Files 63 63
Lines 3700 3700
Branches 524 524
=======================================
Hits 3606 3606
Misses 55 55
Partials 39 39
Flags with carried forward coverage won't be shown. Click here to find out more. Continue to review full report at Codecov.
|
0a2e44a
to
008c773
Compare
a14db89
to
1c29d51
Compare
81745e1
to
ef4339d
Compare
3cd0c0c
to
f061a3f
Compare
5ef0781
to
feda150
Compare
Added a second notebook on errors (the calc in #764 ) and we see compatible errors with various versions of how to calculate them. maybe a question to @alexander-held: is it important to your use-case to have MINUIT-like uncertainties or would uncertainties based on the standard covariance matrix (the blue dashed line) suffice? Separately from this we can implement MINOS-like uncert by doing a line search. |
be7a916
to
6735471
Compare
for more information, see https://pre-commit.ci
In the |
the last entry (brown) is using minuit but supplying the AD gradient. This uses |
Thanks, the strategy was the difference between our setups. I was running with 0.5.4 which does not yet include the switch to strategy 0 if gradients are provided. I am surprised that the difference is this large. I thought the HESSE call would help, but maybe the convergence fails completely? Strange since this likelihood is relatively simple. |
Description
spent some more time with MINUIT code to see how it does the covariance calculation and putting the lessons into a notebook. What I was initially confused by is how it recomputes the full hessian only by transforming the hessian in the internal coordinates. Turns out it uses a property that only holds at the minimum
This is just to document some of these things in order to make full covariances available in non-minuit backends @alexander-held