Add bivariate copula implementation by chrischu12 · Pull Request #137 · simon-hirsch/ondil

chrischu12 · 2025-07-03T08:51:04Z

No description provided.

simon-hirsch

I miss the base class for the copulas, because now all distributions need to implement the param_link parameter. This should be solved by a CopulaMixin class that implements the necessary methods.

src/ondil/base/distribution.py

src/ondil/distributions/bicop_normal.py

src/ondil/distributions/mv_marg_cop.py

src/ondil/distributions/mv_t_low_rank.py

src/ondil/estimators/online_mvdistreg.py

src/ondil/link/copulalinks.py

Copilot

Pull Request Overview

This PR adds support for bivariate copulas by introducing new link functions and multivariate copula distributions, and integrates them into the online multivariate distribution regression estimator.

Introduce FisherZLink and TauToPar in copulalinks.py
Extend base Distribution to accept param_links and update existing distributions
Add BiCopNormal and MarginalCopula classes for Gaussian copula modeling
Integrate copula-based parameter handling into online_mvdistreg

Reviewed Changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/ondil/link/copulalinks.py	New Fisher Z and Kendall‐tau link functions
src/ondil/link/init.py	Export copula link functions
src/ondil/base/distribution.py	Add support for `param_links` in base `Distribution`
src/ondil/distributions/normal.py	Update constructor signature for `param_links`
src/ondil/distributions/mv_t_low_rank.py	Pass `param_links` through to superclass
src/ondil/distributions/mv_normal_chol.py	Pass `param_links` through to superclass
src/ondil/distributions/mv_marg_cop.py	New `MarginalCopula` distribution
src/ondil/distributions/bicop_normal.py	New `BiCopNormal` Gaussian copula distribution
src/ondil/distributions/init.py	Register new copula distributions
src/ondil/estimators/online_mvdistreg.py	Integrate copula option and branching logic into fitting routine

Comments suppressed due to low confidence (3)

src/ondil/distributions/normal.py:47

The param_links parameter should be typed as dict[int, LinkFunction] instead of LinkFunction, and default to an empty dict. Update the annotation to param_links: dict[int, LinkFunction] = {}.

        param_links: LinkFunction = {},

src/ondil/distributions/bicop_normal.py:33

The constructor uses a singular param_link but the base class expects param_links: dict[int, LinkFunction]. Consider renaming to param_links and accepting a mapping keyed by parameter index.

        param_link: LinkFunction = TauToPar(),

src/ondil/estimators/online_mvdistreg.py:690

The code previously scaled X and stored it in X_scaled but now passes the raw X to _outer_fit, bypassing the scaling step. It should likely be X_scaled here.

        self._outer_fit(X=X, y=y, theta=theta)

src/ondil/link/copulalinks.py

src/ondil/estimators/online_mvdistreg.py

src/ondil/distributions/bicop_normal.py

src/ondil/distributions/mv_marg_cop.py

src/ondil/estimators/online_mvdistreg.py

src/ondil/distributions/__init__.py

src/ondil/distributions/bicop_normal.py

simon-hirsch · 2025-09-11T14:10:47Z

@chrischu12 I have changed the base branch as we #134 is merged now.

src/ondil/distributions/bicop_normal.py

src/ondil/distributions/bicop_studentt.py

src/ondil/distributions/mv_marg_cop.py

src/ondil/distributions/normal.py

src/ondil/links/copulalinks.py

src/ondil/links/loglinks.py

src/ondil/links/copulalinks.py

src/ondil/links/loglinks.py

Copilot

Pull Request Overview

Copilot reviewed 18 out of 18 changed files in this pull request and generated 12 comments.

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

src/ondil/links/copulalinks.py

src/ondil/estimators/online_mvdistreg.py

src/ondil/distributions/bicop_studentt.py

src/ondil/distributions/bicop_gumbel.py

src/ondil/base/distribution.py

h --force origin bivariate_copula

src/ondil/__init__.py

src/ondil/distributions/bicop_normal.py

…ants to top

simon-hirsch · 2026-02-27T09:38:56Z

src/ondil/estimators/online_mvdistreg.py

            # If no regularization is allowed, we just return the theta
            return theta

+    def _make_initial_eta(self, theta: np.ndarray):


Should this be on the copula class?

simon-hirsch · 2026-02-27T09:40:06Z

src/ondil/estimators/online_mvdistreg.py

-                theta[a] = self.distribution.set_initial_guess(y, theta[a], p)
-                theta = self._handle_path_regularization(theta=theta, p=p, a=a)
+            # if (inner_iteration == 0) and (outer_iteration == 0) & (a == 0):
+            #     theta[a] = self.distribution.set_initial_guess(y, theta[a], p)


This should not be commented as it is necessary for the multivariate case. Properly switch between copula and MV case in the fitting

simon-hirsch · 2026-02-27T09:40:32Z

src/ondil/estimators/online_mvdistreg.py

-                    dl2_link = self.distribution.cube_to_flat(dl2_link, param=p)
-                    dl2_link = dl2_link[:, k]
+                        else:
+                            eta = self._make_initial_eta(theta)


This is duplicated above

simon-hirsch · 2026-02-27T09:41:04Z

src/ondil/estimators/online_mvdistreg.py

            # If the likelihood is at some point decreasing, we're breaking
            # Hence we need to store previous iteration values:
-            if (inner_iteration > 0) | (outer_iteration > 0):
+            if (inner_iteration == 0) and (outer_iteration == 0):


Duplicated below.

simon-hirsch · 2026-02-27T09:42:52Z

src/ondil/estimators/online_mvdistreg.py

-                            eta_elem, param=p, k=k, d=self.dim_
-                        )
+
+                        if issubclass(self.distribution.__class__, CopulaMixin):


this block should be as eta_to_theta_element on the copula class and then we don't need to import and check for specific copulas on the main estimator class.

simon-hirsch · 2026-02-27T09:45:31Z

src/ondil/estimators/online_mvdistreg.py

+                            param=p,
+                        ).reshape(-1, 1)
+
+                        if isinstance(


Same for this block. Should also be a function on the copula class (the same as above), eta_to_theta_element

simon-hirsch · 2026-02-27T09:47:10Z

src/ondil/estimators/online_mvdistreg.py

                    old_likelihood = self._current_likelihood[a]

            # If the LL is decreasing, we're resetting to the previous iteration
-            if ((outer_iteration > 0) | (inner_iteration > 1)) & decreasing:


why change this to > 0

simon-hirsch · 2026-02-27T09:48:32Z

src/ondil/estimators/online_mvdistreg.py

+                    if issubclass(self.distribution.__class__, CopulaMixin) and p == 0:
+                        out[a][p] = np.tanh(out[a][p] / 2) * (1 - 1e-8)
+                        out[a][p] = self.distribution.param_link_inverse(
+                            out[a][p] * (1 - 1e-8), param=0
+                        ) * (1 - 1e-8)


if issubclass(self.distribution.class, CopulaMixin) and p == 0:
out[a][p] = np.tanh(out[a][p] / 2) * (1 - 1e-8)
out[a][p] = self.distribution.param_link_inverse(
out[a][p] * (1 - 1e-8), param=0
) * (1 - 1e-8)

This pattern is duplicated a few times across the code. Can this be moved in the param_link_inverse for the copula?

simon-hirsch reviewed Jul 3, 2025

View reviewed changes

simon-hirsch requested review from BerriJ and Copilot July 3, 2025 09:13

Copilot AI reviewed Jul 3, 2025

View reviewed changes

src/ondil/link/copulalinks.py Outdated Show resolved Hide resolved

src/ondil/link/copulalinks.py Outdated Show resolved Hide resolved

src/ondil/estimators/online_mvdistreg.py Outdated Show resolved Hide resolved

simon-hirsch mentioned this pull request Jul 3, 2025

Add Copula #136

Closed

simon-hirsch reviewed Jul 4, 2025

View reviewed changes

chrischu12 force-pushed the bivariate_copula branch from cc8a021 to 48f3249 Compare July 14, 2025 14:59

chrischu12 force-pushed the bivariate_copula branch from 63c30eb to f5a3724 Compare July 25, 2025 12:32

simon-hirsch requested changes Jul 26, 2025

View reviewed changes

src/ondil/distributions/__init__.py Show resolved Hide resolved

src/ondil/distributions/bicop_normal.py Outdated Show resolved Hide resolved

src/ondil/distributions/bicop_normal.py Outdated Show resolved Hide resolved

src/ondil/distributions/bicop_normal.py Outdated Show resolved Hide resolved

simon-hirsch changed the base branch from mvdistreg to main September 11, 2025 14:09

simon-hirsch requested changes Sep 13, 2025

View reviewed changes

simon-hirsch reviewed Sep 17, 2025

View reviewed changes

src/ondil/links/loglinks.py Outdated Show resolved Hide resolved

chrischu12 added 10 commits September 18, 2025 13:04

test

48ed465

test

3c8b3d7

Add bivariate copula implementation

4d04d0b

Add bivariate copula implementation II

9ab5d66

Update Bivariate Copula Implementation

ae0920c

Update after rebase

d935d50

delete experiments

1ee2487

minor updates

740e13e

Minor updates

2c4441c

Update bicop_normal and online_mvdistreg with latest changes

ac94ff5

chrischu12 force-pushed the bivariate_copula branch from c403d59 to ac94ff5 Compare September 19, 2025 14:34

chrischu12 added 3 commits September 19, 2025 16:52

{commit_message}

d0dac0a

gumbel;clayton added

fc97498

Adjust bicops

de1e432

simon-hirsch requested a review from Copilot October 8, 2025 07:53

Copilot AI reviewed Oct 8, 2025

View reviewed changes

simon-hirsch reviewed Oct 8, 2025

View reviewed changes

h --force origin bivariate_copula Outdated Show resolved Hide resolved

src/ondil/__init__.py Outdated Show resolved Hide resolved

src/ondil/distributions/bicop_normal.py Outdated Show resolved Hide resolved

simon-hirsch and others added 12 commits December 19, 2025 09:56

Remove commented code, remove loops around vectorized PDF, move const…

fcb8e12

…ants to top

correct scipy.special imports

37475ce

Remove individual function imports from scipy.special

403b825

Remove more unnecessary loops

dac68f8

Move loops

28ea397

Move loops

23c4870

add numba in t copula

6190097

adjust initial guess

f1e2e62

theta reshape

4ebc0dc

make more robust

8b4f50e

add rvs

5124382

add error message

b5207eb

chrischu12 force-pushed the bivariate_copula branch from f958ffd to b5207eb Compare February 19, 2026 15:52

chrischu12 and others added 12 commits February 24, 2026 11:17

adjustments to distributions

77c5e79

simplify code

c0846a0

update links and comment

49cacb7

Merge remote-tracking branch 'upstream/main' into bivariate_copula

3c0f296

correction for lasso

bc39cd3

Add docstrings

3ca051f

Centralize UMIN/UMAX

8a0d30c

Remove unnecessary imports

185330a

Uncomment logging again

e978741

Remove unused parameter k for dampened predictor

bcb2850

remove random 0

136bba4

Uncomment warning again, removed random 0

8b0cf81

simon-hirsch requested changes Feb 27, 2026

View reviewed changes

simon-hirsch added 4 commits February 27, 2026 11:04

Add copula to docs

3c8a344

Add base classes to docs

43c9b00

Exclude copulas from univariate tests

834beed

Fix license headers in copula src files

0d8f89d

Conversation

chrischu12 commented Jul 3, 2025

Uh oh!

simon-hirsch left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

simon-hirsch commented Sep 11, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment