Proposed Numerical Integration Atom #2735

shokonaasti · 2025-02-25T20:55:22Z

Description

Please include a short summary of the change.
Issue link (if applicable):
This PR creates the numerical_integration atom using discretization methods (trapezoidal, Simpson’s rule, and Monte Carlo) that numerically integrates a convex function of some parameters over a given domain. And is implemented as a wrapper around the existing sum atom. Unit test is included in the test_atoms.py file.

Type of change

[x ] New feature (backwards compatible)
New feature (breaking API changes)
Bug fix
Other (Documentation, CI, ...)

Contribution checklist

[x ] Add our license to new files.
[ x] Check that your code adheres to our coding style.
[x ] Write unittests.
[x ] Run the unittests and check that they’re passing.
[] Run the benchmarks to make sure your change doesn’t introduce a regression.

integration atom that can handle both 1D and ND cases

CLAassistant · 2025-02-25T20:55:28Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

SteveDiamond · 2025-02-25T23:56:03Z

cvxpy/atoms/numerical_integration.py

+from cvxpy.atoms.affine.sum import sum as cvx_sum
+
+
+def numerical_integration(f_callable, w, ranges, g_list=None, num_points=100,


all function arguments need types

should be good now!

cvxpy/atoms/numerical_integration.py

shokonaasti

sorry about that!

shokonaasti · 2025-02-26T16:09:08Z

cvxpy/atoms/numerical_integration.py

+from cvxpy.atoms.affine.sum import sum as cvx_sum
+
+
+def numerical_integration(f_callable, w, ranges, g_list=None, num_points=100,


should be good now!

dberkens · 2025-03-07T17:54:12Z

Hey @shokonaasti - did you see the comments on Discord? Copied below for visibility. My sense is maybe we should change it to cp.integrate to be consistent with scipy numerical integration. Any thoughts on the other issues raised?

Hey thanks a lot for your PR.. I had a few comments open for discussion (hence why I didn't give a code review)

I believe we should stick with the scientific python API for these functions: simpson and trapezoid. So it would look something like cp.simpson(expr,...), what do you think?
I personally think passing the cvxpy expression directly (instead of using f_callable on data points) is more in line with our other atoms.. but open to hearing your thoughts on that also
can you add some documentation on the ND integration and the fractional volume grid? Are there numpy equivalent functions or wikipedia links to the algorithm? The code isn't so easy to understand. And it would help to know what they do.
We would need some improved documentation for the internal functions for maintainability
Finally, we need to add the atoms to the webpage.. I can help with that once necessary

shokonaasti · 2025-03-30T22:29:17Z

I made the changes to to adhere to scientific notation as well as passing the cvxpy expression directly. Sorry for the delay and let me know if there are any questions.

PTNobel

I think I'd prefer to have one file intergrate.py that contains both atoms than the subfolder. Any reason to not to do it this way? Are you planning to add a lot more integration methods in future PRs?

Also, just out of lay-curiosity can Forward Euler or Backward Euler be simulated with these atoms?

PTNobel · 2025-04-04T16:01:10Z

cvxpy/atoms/__init__.py

 from cvxpy.atoms.stats import mean, std, var
 from cvxpy.atoms.ptp import ptp
-
+from . import integrate


Suggested change

from . import integrate

from cvxpy.atoms.integrate import trapz, simpson

PTNobel · 2025-04-04T16:01:20Z

cvxpy/atoms/__init__.py

 from cvxpy.atoms.ptp import ptp
-
+from . import integrate
+from cvxpy.atoms import integrate


Suggested change

from cvxpy.atoms import integrate

PTNobel · 2025-04-04T16:02:07Z

cvxpy/atoms/integrate/__init__.py

@@ -0,0 +1,19 @@
+"""
+Copyright, the CVXPY Ashok Viswanathan


Suggested change

Copyright, the CVXPY Ashok Viswanathan

Copyright, the CVXPY Project

Please sign the CLA.

PTNobel · 2025-04-04T16:02:20Z

cvxpy/atoms/integrate/simpson.py

@@ -0,0 +1,97 @@
+"""
+Copyright, the CVXPY Ashok Viswanathan


Suggested change

Copyright, the CVXPY Ashok Viswanathan

Copyright, the CVXPY Project

PTNobel · 2025-04-04T16:03:56Z

cvxpy/atoms/integrate/__init__.py

+from .trapz import trapz
+from .simpson import simpson


Suggested change

from .trapz import trapz

from .simpson import simpson

from cvxpy.atoms.integrate.trapz import trapz

from cvxpy.atoms.integrate.simpson import simpson

PTNobel · 2025-04-04T16:12:48Z

cvxpy/atoms/integrate/simpson.py

+    x: Optional[np.ndarray] = None,
+    dx: float = 1.0,
+    axis: int = -1,
+    even: str = "avg"


Which strategy does scipy implement? I notice that scipy.integrate.simpson doesn't have an even mode specification.

PTNobel · 2025-04-04T16:14:09Z

cvxpy/atoms/integrate/trapz.py

+    axis: int = -1
+) -> Expression:
+    y_ndim = len(y.shape)
+    axis = axis % y_ndim


Same question as with simpson.

PTNobel · 2025-04-04T16:15:16Z

cvxpy/atoms/integrate/trapz.py

+        dxs = np.diff(x)
+        dxs_shape = [1] * y_ndim
+        dxs_shape[axis] = len(dxs)
+        dxs_broadcast = dxs.reshape(dxs_shape)


Should we use np.broadcast_to instead of .reshape

PTNobel · 2025-04-04T16:16:01Z

cvxpy/atoms/integrate/trapz.py

+        center = cvx_sum(y[slicer(1, n - 1)])
+        return dx * (edge + center)
+
+# def trapz(


We generally don't allow commented out code to be merged.

PTNobel · 2025-04-04T16:16:18Z

cvxpy/tests/test_atoms.py

        # The optimized result should be smaller than the naive result,
        # where X of the naive result is I.
        self.assertTrue(prob.value < naiveRes)
+    def test_trapz_atom(self) -> None:


Suggested change

def test_trapz_atom(self) -> None:

def test_trapz_atom(self) -> None:

PTNobel · 2025-06-23T23:57:37Z

@shokonaasti are you able to sign the CLA, so someone else can finish this PR? Will you be able to finish the PR soon?

shokonaasti added 14 commits January 8, 2025 05:41

Add files via upload

9ecd4c1

Update __init__.py

be5db47

Add files via upload

a63b7d8

Update __init__.py

b43e9c5

Update __init__.py

646f850

Add files via upload

c2278e3

Delete cvxpy/atoms/numerical_integration_ND.py

a3010fa

Add files via upload

b2ec4a6

integration atom that can handle both 1D and ND cases

Update __init__.py

cf0a85b

Update numerical_integration.py

44cb160

Update test_atoms.py

16e172f

Delete cvxpy/atoms/numerical_integration_2D.py

7c466be

Delete cvxpy/atoms/numerical_integration_1D.py

f7a64e6

Update numerical_integration.py

14a86bb

SteveDiamond reviewed Feb 25, 2025

View reviewed changes

Update numerical_integration.py

92e6fcb

SteveDiamond reviewed Feb 26, 2025

View reviewed changes

cvxpy/atoms/numerical_integration.py Outdated Show resolved Hide resolved

Update numerical_integration.py

6de5e05

shokonaasti commented Feb 27, 2025

View reviewed changes

shokonaasti added 5 commits March 30, 2025 15:11

Create __init__.py

fe5318c

Add files via upload

a3f78bc

Update __init__.py

8990705

Update test_atoms.py

6a7c30b

Update trapz.py

23cf623

Delete cvxpy/atoms/numerical_integration.py

104a458

PTNobel requested changes Apr 4, 2025

View reviewed changes

		from cvxpy.atoms.affine.sum import sum as cvx_sum


		def numerical_integration(f_callable, w, ranges, g_list=None, num_points=100,

	from . import integrate
	from cvxpy.atoms.integrate import trapz, simpson

	Copyright, the CVXPY Ashok Viswanathan
	Copyright, the CVXPY Project

	def test_trapz_atom(self) -> None:

	def test_trapz_atom(self) -> None:

Proposed Numerical Integration Atom #2735

Are you sure you want to change the base?

Proposed Numerical Integration Atom #2735

Uh oh!

Conversation

shokonaasti commented Feb 25, 2025

Description

Type of change

Uh oh!

CLAassistant commented Feb 25, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

shokonaasti left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dberkens commented Mar 7, 2025

Uh oh!

shokonaasti commented Mar 30, 2025

Uh oh!

PTNobel left a comment

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

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!

PTNobel commented Jun 23, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants