v0.3.7 #12

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.

Sign up for GitHub

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

Merged

lukelowry merged 9 commits into main from dev

Jan 26, 2026

.gitignore

-Original file line number
+Diff line change
@@ Expand Up / @@ -7,3 +7,4 @@ dist @@
     .benchmarks
     benchmark_*.pdf
     examples/.data_dir
+    examples/.data_dir_b

CHANGELOG.rst

-Original file line number
+Diff line change
@@ Expand Up @@
     The format is based on `Keep a Changelog <https://keepachangelog.com/en/1.0.0/>`_,
     and this project adheres to `Semantic Versioning <https://semver.org/spec/v2.0.0.html>`_.
+    [0.3.7] - 2026-01-25
+    --------------------
+    **Changed**
+    - Convolve and DyConvolve can take a scalar argument for scales, and will return transformed signal instead of list in that case.
+    - Convolve and DyConvolve can also take a 1D signal and will return a transformed 1D signal in that case
+    - impulse function now 1D by default
+    - Misc examples updated to show this feature
+    - Broke up benchmark into four seperatea performance tests
+    **Added**
+    - Improved Chebyshev approximator
     [0.3.6] - 2026-01-18
     --------------------
@@ Expand Down Expand Up @@
     **Added**
     - First stable public release of the ``sgwt`` package.
-    - Comprehensive documentation, usage examples, and testing suite.
+    - Comprehensive documentation, usage examples, and testing suite.

docs/_static/images/demo_benchmark_a.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_benchmark_b.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_benchmark_c.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_benchmark_d.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_cheby_coeffs.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_cheby_convolve.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_cheby_convolve_2.png

Binary file not shown.

docs/_static/images/demo_cheby_time.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/_static/images/demo_mesh_wavelet_1.png

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

docs/examples/benchmarks.rst

-Original file line number
+Diff line change
@@ Expand Up @@
        :maxdepth: 1
        :caption: Analytical Filter Performance
-       benchmarks/demo_benchmark
+       benchmarks/demo_benchmark_a
+       benchmarks/demo_benchmark_b
+       benchmarks/demo_benchmark_c
+       benchmarks/demo_benchmark_d
     .. toctree::
        :maxdepth: 1
@@ Expand Down @@

docs/examples/benchmarks/demo_benchmark.rst

This file was deleted.

docs/examples/benchmarks/demo_benchmark_a.rst

-Original file line number
+Diff line change
@@ -0,0 +1,9 @@
+    Benchmark: Graph Size Scaling
+    =============================
+    This benchmark shows how execution time scales with graph size (number of edges) for both ``Convolve`` (Static) and ``DyConvolve`` (Dynamic) solvers across lowpass, bandpass, and highpass filters.
+    .. image:: /_static/images/demo_benchmark_a.png
+       :alt: Graph Size Scaling Benchmark
+       :align: center
+       :width: 80%

docs/examples/benchmarks/demo_benchmark_b.rst

-Original file line number
+Diff line change
@@ -0,0 +1,9 @@
+    Benchmark: Signal Count Scaling
+    ===============================
+    This benchmark shows how execution time scales with the number of input signals (M) for the analytical graph wavelet solvers.
+    .. image:: /_static/images/demo_benchmark_b.png
+       :alt: Signal Count Scaling Benchmark
+       :align: center
+       :width: 80%

docs/examples/benchmarks/demo_benchmark_c.rst

-Original file line number
+Diff line change
@@ -0,0 +1,9 @@
+    Benchmark: Scale Count Scaling
+    ==============================
+    This benchmark shows how execution time scales with the number of wavelet scales (J) for the analytical graph wavelet solvers.
+    .. image:: /_static/images/demo_benchmark_c.png
+       :alt: Scale Count Scaling Benchmark
+       :align: center
+       :width: 80%

docs/examples/benchmarks/demo_benchmark_d.rst

-Original file line number
+Diff line change
@@ -0,0 +1,9 @@
+    Benchmark: Filter Order Scaling
+    ===============================
+    This benchmark shows how bandpass filter order (K) affects execution time for the analytical graph wavelet solvers.
+    .. image:: /_static/images/demo_benchmark_d.png
+       :alt: Filter Order Scaling Benchmark
+       :align: center
+       :width: 80%

docs/usage/input_signals.rst

            
                      Original file line number
                      Diff line number
                      Diff line change
                  
    @@ -6,27 +6,34 @@ The primary input for all filtering operations is a signal defined on the vertic
  
    Format and Memory Layout

    ------------------------

    The signal array ``X`` should have the shape ``(n_vertices, n_timesteps)``, where:

    The signal array ``X`` can be either:

    - ``n_vertices`` is the number of vertices (nodes) in the graph, matching ``L.shape[0]``.

    - ``n_timesteps`` is the number of independent signals or time samples to be processed concurrently.

    - **1D**: Shape ``(n_vertices,)`` for a single signal

    - **2D**: Shape ``(n_vertices, n_timesteps)`` for multiple signals

    For optimal performance, it is highly recommended to create the signal array with **column-major (Fortran) ordering** by specifying ``order='F'``. This memory layout aligns with the underlying C-based ``CHOLMOD`` library, avoiding costly data re-ordering during computation.

    where ``n_vertices`` is the number of vertices (nodes) in the graph, matching ``L.shape[0]``, and ``n_timesteps`` is the number of independent signals or time samples to be processed concurrently.

    When a 1D signal is passed, the output will also be 1D (squeezed accordingly).

    For optimal performance with 2D signals, it is highly recommended to create the signal array with **column-major (Fortran) ordering** by specifying ``order='F'``. This memory layout aligns with the underlying C-based ``CHOLMOD`` library, avoiding costly data re-ordering during computation.

    .. code-block:: python

        import numpy as np

        from sgwt import DELAY_USA as L

        n_vertices = L.shape[0]

        n_signals = 10

        # Create a random signal with the required ordering

        X = np.random.randn(n_vertices, n_signals).astype(np.float64)

        X = np.asfortranarray(X)

        # 1D signal (single signal)

        X_1d = np.random.randn(n_vertices)

        # 2D signal (multiple signals)

        n_signals = 10

        X_2d = np.random.randn(n_vertices, n_signals).astype(np.float64)

        X_2d = np.asfortranarray(X_2d)

        # Verify the memory order

        assert X.flags['F_CONTIGUOUS']

        # Verify the memory order for 2D

        assert X_2d.flags['F_CONTIGUOUS']

    .. seealso::

       :doc:`../theory/theory_graph`

docs/usage/quickstart.rst

-Original file line number
+Diff line change
@@ Expand Up / @@ -35,6 +35,7 @@ What's Happening in This Example? @@
 .  **Define Filter Parameters**
         - ``scales = [0.1, 1.0, 10.0]``: We specify the scales for our band-pass filter. In spectral graph filtering, the scale is inversely related to frequency. Smaller scales target high-frequency components (details, sharp changes), while larger scales target low-frequency components (smooth variations).
+        - You can also pass a single scalar value (e.g., ``scales=1.0``), in which case the output will be a single array instead of a list.
 .  **Perform the Convolution**
         - ``with Convolve(L) as conv:``: This is the core of the library. We create a :class:`~sgwt.Convolve` context. Upon entry, it performs an efficient one-time symbolic factorization of the graph Laplacian ``L``. This pre-computation makes all subsequent filtering operations extremely fast. The context also manages all the low-level memory required by the CHOLMOD backend.
@@ Expand Down @@