Temperature feature

[jackraymond]

Add generators for the background susceptibility correction to binary quadratic models.
Generalize the maximum_pseudolikelihood estimator for multidimensional parameter inference: use cases include estimation of the background susceptibility, estimation of a conversion factor from flux biases to (unitless) linear fields, estimating relative rescaling of h at freezeout when using h_gain.
Add type hinting, improve documentation.
Standardize formatting (black) on temperature.py and related tests.

[jackraymond]

@LoganLim2271 has offered to review.
Seems like my environment is not standard since some tests are failing. Will correct those now.

[jackraymond]

Can't reproduce these failing tests in a new python 3.13 environment. The errors relate to SciPy return object, but scipy 1.15 (default install) and mos trecent 2.2.1 both throw no errors for me.

[codecov]

Codecov Report

Attention: Patch coverage is 84.38819% with 37 lines in your changes missing coverage. Please review.

Project coverage is 85.81%. Comparing base (1bb84c6) to head (9a60c0b).

Files with missing lines	Patch %	Lines
dwave/system/temperatures.py	84.38%	37 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #552      +/-   ##
==========================================
- Coverage   88.75%   85.81%   -2.94%     
==========================================
  Files          21       21              
  Lines        1654     1777     +123     
==========================================
+ Hits         1468     1525      +57     
- Misses        186      252      +66     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jackraymond]

@kevinchern is also willing to take a look

[LoganLim2271]

Have to look at this intermittently, some minor comments

[kevinchern]

Some to-dos

1. typeset: math, URLs, variable/code name
2. complete docstrings, e.g., args and return types

What are the motivation for the many customizable arguments? For example,

1. passing in degenerative_fields
2. optimization method (is there a robust method)
3. univariate case versus calling multivariate case with, e.g., f(x)->f([x])

[kevinchern]

The default excludes 0.
Should note the following in the docstring:

1. The arbitrary choice of 0.001 and
2. There is a check for when T=0

[kevinchern]

don't the optimization methods in scipy return boundary values? i.e., no need to check for T=0 and leave it to scipy to return 0

[jackraymond]

I don't trust scipy to handle the case beta=Infinity (T=0) correctly. It turns out this is a very common outcome since the main usage is for heuristic optimizers that return local minima exclusively (or disproportionately).

[kevinchern]

same comment as above: method-specific args could be packed into a kwargs_opt=None arg

[jackraymond]

This is a good suggestion. For legacy reasons I didn't think of it (was originally matlab translated to python within sciPy dependence).

[kevinchern]

Document this special case

[kevinchern]

:D

[thisac]

This has become a very long function with many branches. Could it be split up and organized into smaller functions for the various cases and checks? That would make it much easier to review and maintain, and would also allow for better documentation and comments.

[jackraymond]

Some to-dos

1. typeset: math, URLs, variable/code name

2. complete docstrings, e.g., args and return types

What are the motivation for the many customizable arguments? For example,

1. passing in `degenerative_fields`

2. optimization method (is there a robust method)

3. univariate case versus calling multivariate case with, e.g.,  f(x)->f([x])

1. This is a recent addition. Could hardcode True, will think about this.
2. For well posed problems all methods are robust, convex optimization. I should probably take greater advantage of this to simplify the code. However, need to better document the catching/special treatment of pathological cases.
3. Called functions in numpy and SciPy rely on the same distinction is the main one. Its hard work to continually turn scalars into lists.