_PyfarMultichannel class by h-chmeruk · Pull Request #837 · pyfar/pyfar

h-chmeruk · 2025-09-01T09:41:38Z

As proposed in the Issue #835, here is a draft implementation of the _NumPyBase class. This abstract class should define the NumPy-like functions that must then be implemented by the _Audio, Coordinates and other classes.

Edit: From the fourth commit , the class _NumPyBase is now called _PyfarMultichannel.

sikersten

Thank you for getting this started!
To keep it simple, please create a new pull request for every new method and only contain existing methods in this pull request.

pyfar/classes/numpy_like.py

…ded comment.setter

ahms5

looks already very nice, thank you.

pyfar/classes/_pyfar_multichannel.py

ahms5 · 2025-09-02T16:00:06Z

pyfar/classes/_pyfar_multichannel.py

+    @abstractmethod
+    def flatten(self):
+        """Return a copy of the object collapsed into one channel dimension."""
+
+        pass
+    @abstractmethod
+    def transpose(self):
+        """Returns a copy of the object with channel axes transposed."""
+
+        pass


i think flatten and transpose can also be implemented based on reshape, see _Audio class as an example

works for flatten, not sure about transpose

Transpose is quite extensive for _Audio and not defined for other classes (please correct me if I'm wrong), so I would leave it as an abstract method...

ahms5 · 2025-09-02T16:03:25Z

pyfar/classes/_pyfar_multichannel.py

+    @abstractmethod
+    def broadcast_cdim(self, cdim):
+        """
+        Broadcast an copy of the object wth a certain channel dimension
+        (`cdim`).
+
+        Parameters
+        ----------
+        cdim : int
+            The cdim to which the object is broadcasted.
+
+        Returns
+        -------
+        object : _PyfarMultichannel
+            Broadcasted copy of the object.
+        """
+
+        pass


other remark, based on what broadcast_cdim does, i would wonder if this function really make sence, no such numpy method exists.
inserting new dimension can also be done via e.g. reshape

Agree! It is essentially a convenience function to prepend dimensions. I think we can leave it if properly documented.

ahms5 · 2025-09-02T16:04:03Z

pyfar/classes/_pyfar_multichannel.py

+    @abstractmethod
+    def cshape(self):
+        """
+        Return channel shape.


extend doc, similar to cdim

I'm not sure what the docs should state here :P

I'm not sure how to extend the description of cshape, as it returns the shape excluding one dimension for _Audio and Coordinates, and does not yet exist for Orientations and Filters. There is a cshape for TransmissionMatrix, as it inherits from FrequencyData, and there is also abcd_cshape, which excludes two dimensions from the cshape.

e.g. like this: https://pyfar.readthedocs.io/en/stable/classes/pyfar.audio.html#pyfar.FrequencyData.cshape

pyfar/classes/_pyfar_multichannel.py

ahms5 · 2025-09-02T16:17:24Z

i have just noticed, that __getitem__ and __setitem__ are not there yet, this is the implentation for slicing, see eg audio classes

…improved docstrings

pyfar/classes/_pyfar_multichannel.py

sikersten · 2025-09-03T08:43:23Z

pyfar/classes/_pyfar_multichannel.py

+    @abstractmethod
+    def flatten(self):
+        """Return a copy of the object collapsed into one channel dimension."""
+
+        pass
+    @abstractmethod
+    def transpose(self):
+        """Returns a copy of the object with channel axes transposed."""
+
+        pass


works for flatten, not sure about transpose

pyfar/classes/_pyfar_multichannel.py

sikersten · 2025-09-03T08:49:58Z

pyfar/classes/_pyfar_multichannel.py

+    @abstractmethod
+    def broadcast_cdim(self, cdim):
+        """
+        Broadcast an copy of the object wth a certain channel dimension
+        (`cdim`).
+
+        Parameters
+        ----------
+        cdim : int
+            The cdim to which the object is broadcasted.
+
+        Returns
+        -------
+        object : _PyfarMultichannel
+            Broadcasted copy of the object.
+        """
+
+        pass


Agree! It is essentially a convenience function to prepend dimensions. I think we can leave it if properly documented.

pyfar/classes/_pyfar_multichannel.py

…sses/PyfarBase

ahms5

I would suggest to assume that _data contains the data for defuld behaviour. And we could still add tests.

pyfar/classes/_pyfar_multichannel.py

ahms5 · 2025-10-08T10:03:04Z

pyfar/classes/_pyfar_multichannel.py

+    def transpose(self, caxes=None):
+        """Returns a copy of the object with channel axes transposed."""
+
+        pass


ahms5 · 2025-10-08T10:03:16Z

pyfar/classes/_pyfar_multichannel.py

+            Broadcasted copy of the object.
+        """
+
+        pass


ahms5 · 2025-10-08T10:07:39Z

pyfar/classes/_pyfar_multichannel.py

if the fle jsut contains one class i would rename it to _PyfarMultichannel.py

pyfar/classes/_pyfar_multichannel.py

…ultichannel, Improved docstrings

…im, added T(self) property, improved docstrings

ahms5

Thank you Hanna, looks almost ready imo. I just have small comments on tests

ahms5 · 2025-10-17T10:28:15Z

pyfar/classes/__init__.py

+    '_pyfar_multichannel',
+


Suggested change

'_pyfar_multichannel',

ahms5 · 2025-10-17T10:29:17Z

pyfar/classes/__init__.py

 from . import filter
 from . import transmission_matrix
 from . import warnings
+from . import _pyfar_multichannel


Suggested change

from . import _pyfar_multichannel

abstract classes are not part of the official pyfar api. you can still acess them via pyfar.classes._pyfar_multichannel._PyfarMultichannel

ahms5 · 2026-01-21T17:08:36Z

pyfar/classes/_pyfar_multichannel.py

+from math import prod
+from pyfar.classes._pyfar_base import _PyfarBase
+
+class _PyfarMultichannel(_PyfarBase, ABC):


Suggested change

class _PyfarMultichannel(_PyfarBase, ABC):

class _PyfarMultichannel(_PyfarBase):

_PyfarBase already inherit from ABC

ahms5 · 2026-01-21T17:12:25Z

pyfar/classes/_pyfar_multichannel.py

+            Broadcasted copy of the object.
+        """
+
+        signal = self.copy()


here the input check is missing

ahms5 · 2026-01-21T17:14:25Z

tests/test_pyfar_multichannel.py

+@patch.multiple(_PyfarMultichannel, __abstractmethods__=set())
+def test_cshape():
+    """Test the cshape property."""
+    data = np.array([[[1, 1, 1], [1, 1, 1]], [[2, 2, 2], [2, 2, 2]]])


Suggested change

data = np.array([[[1, 1, 1], [1, 1, 1]], [[2, 2, 2], [2, 2, 2]]])

data = np.ones((2,2,2))

this would be much easier to read

ahms5 · 2026-01-21T17:15:58Z

tests/test_pyfar_multichannel.py

+@pytest.mark.parametrize('cshape', [(5, 1), (2, 3)])
+def test_reshape_value_error(cshape):
+    """Test whether ValueError is raised for the reshape method."""
+    data = np.array([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]])


ahms5 · 2026-01-21T17:16:26Z

tests/test_pyfar_multichannel.py

+@patch.multiple(_PyfarMultichannel, __abstractmethods__=set())
+def test_flatten():
+    """Test the flatten method."""
+    data = np.array([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]])


ahms5 · 2026-01-21T17:18:29Z

tests/test_pyfar_multichannel.py

+    rng = np.random.default_rng()
+    data = rng.random((6, 2, 5, 256))


Suggested change

rng = np.random.default_rng()

data = rng.random((6, 2, 5, 256))

data = np.arange(2*3*4).reshape((1, 2, 3, 4))

better not to use randomized values in tests. tests should be determistic

ahms5 · 2026-01-21T17:19:45Z

tests/test_pyfar_multichannel.py

+@pytest.mark.parametrize('taxis', [(2, 0, 1), (-1, 0, -2)])
+def test_transpose_positional_arguments(taxis):
+    rng = np.random.default_rng()
+    data = rng.random((6, 2, 5, 256))


ahms5 · 2026-01-21T17:19:51Z

tests/test_pyfar_multichannel.py

+@patch.multiple(_PyfarMultichannel, __abstractmethods__=set())
+def test_transpose_errors():
+    rng = np.random.default_rng()
+    data = rng.random((10, 10, 256))


h-chmeruk · 2026-01-22T16:13:02Z

pyfar/classes/_pyfar_multichannel.py

+        cshape : tuple, list, np.ndarray
+            The channel shape to which the object is broadcasted.


I'm not entirely sure, is it a good idea to allow all these data types for cshape?

First draft of the _PyfarBase class

3ffb0bc

h-chmeruk added audio Classes pyfar.classes.audio coordinates labels Sep 1, 2025

sikersten requested review from ahms5 and sikersten September 1, 2025 12:04

sikersten requested changes Sep 1, 2025

View reviewed changes

Renamed file pyfar_base.py, removed channel-related functionality, ad…

4804ecc

…ded comment.setter

h-chmeruk requested a review from sikersten September 2, 2025 11:33

h-chmeruk changed the title ~~First draft of the _NumPyLike class~~ First draft of the _PyfarMultichannel class Sep 2, 2025

ahms5 requested changes Sep 2, 2025

View reviewed changes

ahms5 added this to the v0.9.0 milestone Sep 2, 2025

ahms5 assigned h-chmeruk Sep 2, 2025

ahms5 added this to Weekly Planning Sep 2, 2025

github-project-automation bot moved this to Backlog in Weekly Planning Sep 2, 2025

ahms5 moved this from Backlog to Drafting Phase in Weekly Planning Sep 2, 2025

sikersten changed the base branch from develop to enhance/base_classes September 3, 2025 08:11

sikersten mentioned this pull request Sep 3, 2025

Add base classes for pyfar functionality #838

Closed

h-chmeruk added 2 commits September 3, 2025 10:28

Added default implementation for the _encode method

668b00d

Deleted old pyfar_base.py, added default implementation for _encode, …

d8fcfb4

…improved docstrings

sikersten requested changes Sep 3, 2025

View reviewed changes

github-project-automation bot moved this from Drafting Phase to Require review in Weekly Planning Sep 3, 2025

sikersten requested review from f-brinkmann, hoyer-a and mberz September 3, 2025 09:06

ahms5 moved this from Require review to Open Discussion in Weekly Planning Sep 4, 2025

h-chmeruk added 2 commits September 4, 2025 16:19

Added minor change in the _encode function

0381b4d

Added test functions for the _PyfarBase class

005e8ba

ahms5 requested review from ahms5 and sikersten September 9, 2025 12:32

h-chmeruk added 2 commits September 29, 2025 15:31

Renamed files, added other small changes

806b658

Merge branch 'develop' of https://github.com/pyfar/pyfar into BaseCla…

954ab1e

…sses/PyfarBase

ahms5 requested changes Oct 8, 2025

View reviewed changes

github-project-automation bot moved this from Open Discussion to Require review in Weekly Planning Oct 8, 2025

ahms5 changed the base branch from enhance/base_classes to develop_base_classes October 10, 2025 14:35

ahms5 mentioned this pull request Oct 10, 2025

Collection for the new BaseClass features #850

Draft

4 tasks

ahms5 added the base_classes label Oct 10, 2025

sikersten requested a review from artpelling October 14, 2025 07:35

ahms5 reviewed Oct 17, 2025

View reviewed changes

pyfar/classes/_pyfar_multichannel.py Outdated Show resolved Hide resolved

h-chmeruk added 8 commits October 24, 2025 08:57

Replaced the Dummy class by patching __abstractmethods__, renamed files

a17fd77

Removed blank lines after one-line docstrings

256080b

First draft of the _NumPyLike class

63b40ea

Removed new methods from this PR, class _NumPyLike renamed to _PyfarM…

7aa3e27

…ultichannel, Improved docstrings

Added setter and getter methods, implemented flatten and broadcast_cd…

6a6e031

…im, added T(self) property, improved docstrings

Draft version of the testing functions

354dee9

Branch rebased to BaseClasses/PyfarBase, extended test functions

171526d

Added blank lines between tests in the test_pyfar_multichannel.py file

ee2cfa7

h-chmeruk force-pushed the BaseClasses/NumPyLike branch from f138bd7 to ee2cfa7 Compare November 18, 2025 14:34

h-chmeruk changed the title ~~First draft of the _PyfarMultichannel class~~ _PyfarMultichannel class Nov 18, 2025

h-chmeruk changed the base branch from develop_base_classes to BaseClasses/PyfarBase November 18, 2025 14:35

h-chmeruk marked this pull request as ready for review November 18, 2025 14:36

h-chmeruk requested a review from ahms5 November 18, 2025 14:36

Base automatically changed from BaseClasses/PyfarBase to develop_base_classes November 28, 2025 15:23

Added implementation for the transpose() function, added test functions

33bd1d2

ahms5 requested changes Jan 21, 2026

View reviewed changes

Applied suggestions from code review by @ahms5

ba04400

h-chmeruk commented Jan 22, 2026

View reviewed changes

h-chmeruk requested a review from ahms5 February 4, 2026 09:54

mberz moved this from Require review to Implementation in progress in Weekly Planning Mar 18, 2026

ahms5 force-pushed the develop_base_classes branch from 9c5d755 to 796276a Compare April 1, 2026 07:02

	class _PyfarMultichannel(_PyfarBase, ABC):
	class _PyfarMultichannel(_PyfarBase):

	data = np.array([[[1, 1, 1], [1, 1, 1]], [[2, 2, 2], [2, 2, 2]]])
	data = np.ones((2,2,2))

		rng = np.random.default_rng()
		data = rng.random((6, 2, 5, 256))

	rng = np.random.default_rng()
	data = rng.random((6, 2, 5, 256))
	data = np.arange(234).reshape((1, 2, 3, 4))

		cshape : tuple, list, np.ndarray
		The channel shape to which the object is broadcasted.

Conversation

h-chmeruk commented Sep 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sikersten 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!

ahms5 left a comment

Choose a reason for hiding this comment

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!

sikersten Sep 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

Uh oh!

ahms5 commented Sep 2, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

sikersten Sep 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

ahms5 left a comment

Choose a reason for hiding this comment

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!

Uh oh!

ahms5 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!

h-chmeruk commented Sep 1, 2025 •

edited

Loading

sikersten Sep 3, 2025 •

edited

Loading

sikersten Sep 3, 2025 •

edited

Loading