[DOC] Add implementation references (#2748)

* implementation references

* better attribution

Author: MatthewMiddlehurst

aeon/base/_compose.py

@@ -12,8 +12,10 @@
 class ComposableEstimatorMixin(ABC):
     """Handles parameter management for estimators composed of named estimators.
 
-    Parts (i.e. get_params and set_params) adapted or copied from the scikit-learn
+    Parts (i.e. get_params and set_params) adapted from the scikit-learn 1.5.0
     ``_BaseComposition`` class in utils/metaestimators.py.
+    https://github.com/scikit-learn/scikit-learn/
+    Copyright (c) 2007-2024 The scikit-learn developers, BSD-3
     """
 
     # Attribute name containing an iterable of processed (str, estimator) tuples

aeon/classification/dictionary_based/_redcomets.py

@@ -66,7 +66,8 @@ class REDCOMETS(BaseClassifier):
 
     Notes
     -----
-    Adapted from the implementation at https://github.com/zy18811/RED-CoMETS
+    Adapted from the implementation at https://github.com/zy18811/RED-CoMETS by
+    the code owner.
 
     References
     ----------

aeon/clustering/_kernel_k_means.py

@@ -64,7 +64,7 @@ def _kdtw_lk(x, y, local_kernel):
     return cost_matrix[x_timepoints - 1, y_timepoints - 1]
 
 
-def kdtw(x, y, sigma=1.0, epsilon=1e-3):
+def _kdtw(x, y, sigma=1.0, epsilon=1e-3):
     """
     Callable kernel function for KernelKMeans.
 
@@ -81,6 +81,12 @@ def kdtw(x, y, sigma=1.0, epsilon=1e-3):
         A small constant added for numerical stability to avoid zero values in the
         local kernel matrix.
 
+    Notes
+    -----
+    Inspired by the original implementation
+    https://github.com/pfmarteau/KDTW/tree/master
+    Copyright (c) 2020 Pierre-François Marteau, MIT License
+
     Returns
     -------
     similarity : float
@@ -92,35 +98,6 @@ def kdtw(x, y, sigma=1.0, epsilon=1e-3):
     return _kdtw_lk(x, y, local_kernel)
 
 
-def factory_kdtw_kernel(channels: int):
-    """
-    Return a kdtw kernel callable function that flattened samples to (T, channels).
-
-    Parameters
-    ----------
-    channels: int
-        Number of channels per timepoint.
-
-    Returns
-    -------
-    kdtw_kernel : callable
-        A callable kernel function that computes the KDTW similarity between two
-        time series samples. The function signature is the same as the kdtw
-        function.
-    """
-
-    def kdtw_kernel(x, y, sigma=1.0, epsilon=1e-3):
-        if x.ndim == 1:
-            T = x.size // channels
-            x = x.reshape(T, channels)
-        if y.ndim == 1:
-            T = y.size // channels
-            y = y.reshape(T, channels)
-        return kdtw(x, y, sigma=sigma, epsilon=epsilon)
-
-    return kdtw_kernel
-
-
 class TimeSeriesKernelKMeans(BaseClusterer):
     """Kernel K Means [1]_: wrapper of the ``tslearn`` implementation.
 
@@ -255,7 +232,18 @@ def _fit(self, X, y=None):
             verbose = 1
 
         if self.kernel == "kdtw":
-            self.kernel = factory_kdtw_kernel(channels=X.shape[1])
+            n_channels = X.shape[1]
+
+            def kdtw_kernel(x, y, sigma=1.0, epsilon=1e-3):
+                if x.ndim == 1:
+                    T = x.size // n_channels
+                    x = x.reshape(T, n_channels)
+                if y.ndim == 1:
+                    T = y.size // n_channels
+                    y = y.reshape(T, n_channels)
+                return _kdtw(x, y, sigma=sigma, epsilon=epsilon)
+
+            self.kernel = kdtw_kernel
 
         self._tslearn_kernel_k_means = TsLearnKernelKMeans(
             n_clusters=self.n_clusters,

aeon/distances/elastic/_bounding_matrix.py

@@ -63,8 +63,14 @@ def create_bounding_matrix(
 def _itakura_parallelogram(x_size: int, y_size: int, max_slope_percent: float):
     """Itakura parallelogram bounding matrix.
 
-    This code was adapted from pyts. This link to the original code:
+    This code was adapted from the tslearn and pyts functions.
+
+    pyts code:
     https://pyts.readthedocs.io/en/latest/_modules/pyts/metrics/dtw.html#itakura_parallelogram
+    Copyright (c) 2018, Johann Faouzi and pyts contributors, BSD-3
+    tslearn code (line 974):
+    https://github.com/tslearn-team/tslearn/blob/main/tslearn/metrics/dtw_variants.py
+    Copyright (c) 2017, Romain Tavenard, BSD-2
     """
     one_percent = min(x_size, y_size) / 100
     max_slope = math.floor((max_slope_percent * one_percent) * 100)

aeon/regression/deep_learning/_lite_time.py

@@ -105,6 +105,7 @@ class LITETimeRegressor(BaseRegressor):
     -----
     Adapted from the implementation from Ismail-Fawaz et. al
     https://github.com/MSD-IRIMAS/LITE
+    by the code owner.
 
     References
     ----------
@@ -388,6 +389,7 @@ class IndividualLITERegressor(BaseDeepRegressor):
     -----
     Adapted from the implementation from Ismail-Fawaz et. al
     https://github.com/MSD-IRIMAS/LITE
+    by the code owner.
 
     References
     ----------

aeon/segmentation/_ggs.py

@@ -23,6 +23,7 @@
 Based on the work from [1]_.
 
 - source code adapted based on: https://github.com/cvxgrp/GGS
+  Copyright (c) 2018, Stanford University Convex Optimization Group, BSD-2
 - paper available at: https://stanford.edu/~boyd/papers/pdf/ggs.pdf
 
 References

aeon/transformations/collection/convolution_based/_minirocket.py

@@ -55,7 +55,7 @@ class MiniRocket(BaseCollectionTransformer):
     Notes
     -----
      Directly adapted from the original implementation
-     https://github.com/angus924/minirocket.
+     https://github.com/angus924/minirocket with owner permission.
 
     Examples
     --------

aeon/transformations/series/_bkfilter.py

@@ -34,8 +34,9 @@ class BKFilter(BaseSeriesTransformer):
 
     Notes
     -----
-    Adapted from statsmodels implementation
+    Adapted from statsmodels 0.14.4 implementation
     https://github.com/statsmodels/statsmodels/blob/main/statsmodels/tsa/filters/bk_filter.py
+    Copyright (c) 2009-2018 statsmodels Developers, BSD-3
 
     References
     ----------

aeon/utils/show_versions.py

@@ -1,7 +1,4 @@
-"""Utility methods to print system info for debugging.
-
-Adapted from the sklearn show_versions function.
-"""
+"""Utility methods to print system info for debugging."""
 
 __maintainer__ = ["MatthewMiddlehurst"]
 __all__ = ["show_versions"]
@@ -37,6 +34,12 @@ def show_versions(as_str: bool = False) -> Union[str, None]:
     str or None
         The output string if `as_str` is True, otherwise None.
 
+    Notes
+    -----
+    Adapted from the scikit-learn 1.5.0 show_versions function.
+    https://github.com/scikit-learn/scikit-learn/
+    Copyright (c) 2007-2024 The scikit-learn developers, BSD-3
+
     Examples
     --------
     >>> from aeon.utils import show_versions

aeon/visualisation/results/_critical_difference.py

@@ -88,8 +88,8 @@ def plot_critical_difference(
     overall performance in general, and such comparisons should be seen as exploratory
     analysis rather than designed experiments to test an a priori hypothesis.
 
-    Parts of the code are adapted from here:
-    https://github.com/hfawaz/cd-diagram
+    Parts of the code are adapted from https://github.com/hfawaz/cd-diagram
+    with permission from the owner.
 
     Parameters
     ----------