Support for unequal length timeseries in itakura parallelogram (aeon-toolkit#2647)

Author: tanishy7777

aeon/distances/elastic/_bounding_matrix.py

@@ -63,44 +63,61 @@ 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 tslearn. This link to the original code line 974:
-    https://github.com/tslearn-team/tslearn/blob/main/tslearn/metrics/dtw_variants.py
+    This code was adapted from pyts. This link to the original code:
+    https://pyts.readthedocs.io/en/latest/_modules/pyts/metrics/dtw.html#itakura_parallelogram
     """
-    if x_size != y_size:
-        raise ValueError(
-            """Itakura parallelogram does not support unequal length time series.
-Please consider using a full bounding matrix or a sakoe chiba bounding matrix
-instead."""
-        )
     one_percent = min(x_size, y_size) / 100
     max_slope = math.floor((max_slope_percent * one_percent) * 100)
     min_slope = 1 / float(max_slope)
-    max_slope *= float(x_size) / float(y_size)
-    min_slope *= float(x_size) / float(y_size)
-
-    lower_bound = np.empty((2, y_size))
-    lower_bound[0] = min_slope * np.arange(y_size)
-    lower_bound[1] = (
-        (x_size - 1) - max_slope * (y_size - 1) + max_slope * np.arange(y_size)
-    )
-    lower_bound_ = np.empty(y_size)
-    for i in range(y_size):
-        lower_bound_[i] = max(round(lower_bound[0, i], 2), round(lower_bound[1, i], 2))
-    lower_bound_ = np.ceil(lower_bound_)
-
-    upper_bound = np.empty((2, y_size))
-    upper_bound[0] = max_slope * np.arange(y_size)
-    upper_bound[1] = (
-        (x_size - 1) - min_slope * (y_size - 1) + min_slope * np.arange(y_size)
-    )
-    upper_bound_ = np.empty(y_size)
-    for i in range(y_size):
-        upper_bound_[i] = min(round(upper_bound[0, i], 2), round(upper_bound[1, i], 2))
-    upper_bound_ = np.floor(upper_bound_ + 1)
-
-    bounding_matrix = np.full((x_size, y_size), False)
-    for i in range(y_size):
-        bounding_matrix[int(lower_bound_[i]) : int(upper_bound_[i]), i] = True
+    max_slope *= float(y_size - 1) / float(x_size - 2)
+    max_slope = max(max_slope, 1.0)
+
+    min_slope *= float(y_size - 2) / float(x_size - 1)
+    min_slope = min(min_slope, 1.0)
+
+    centered_scale = np.arange(x_size) - x_size + 1
+
+    lower_bound = np.empty(x_size, dtype=np.float64)
+    upper_bound = np.empty(x_size, dtype=np.float64)
+
+    for i in range(x_size):
+        lb0 = min_slope * i
+        lb1 = max_slope * centered_scale[i] + y_size - 1
+        lower_bound[i] = math.ceil(max(round(lb0, 2), round(lb1, 2)))
+
+        ub0 = max_slope * i + 1
+        ub1 = min_slope * centered_scale[i] + y_size
+        upper_bound[i] = math.floor(min(round(ub0, 2), round(ub1, 2)))
+
+    if max_slope == 1.0:
+        if y_size > x_size:
+            for i in range(x_size - 1):
+                upper_bound[i] = lower_bound[i + 1]
+        else:
+            for i in range(x_size):
+                upper_bound[i] = lower_bound[i] + 1
+
+    for i in range(x_size):
+        if lower_bound[i] < 0:
+            lower_bound[i] = 0
+        if lower_bound[i] > y_size:
+            lower_bound[i] = y_size
+        if upper_bound[i] < 0:
+            upper_bound[i] = 0
+        if upper_bound[i] > y_size:
+            upper_bound[i] = y_size
+
+    bounding_matrix = np.empty((x_size, y_size), dtype=np.bool_)
+    for i in range(x_size):
+        for j in range(y_size):
+            bounding_matrix[i, j] = False
+
+    for i in range(x_size):
+        start = int(lower_bound[i])
+        end = int(upper_bound[i])
+        for j in range(start, end):
+            bounding_matrix[i, j] = True
+
     return bounding_matrix
 
 

aeon/distances/elastic/tests/test_bounding.py

@@ -1,7 +1,6 @@
 """Test for bounding matrix."""
 
 import numpy as np
-import pytest
 
 from aeon.distances import create_bounding_matrix
 
@@ -37,10 +36,34 @@ def test_itakura_parallelogram():
     matrix = create_bounding_matrix(10, 10, itakura_max_slope=0.2)
     assert isinstance(matrix, np.ndarray)
 
-    with pytest.raises(
-        ValueError,
-        match="""Itakura parallelogram does not support unequal length time series.
-Please consider using a full bounding matrix or a sakoe chiba bounding matrix
-instead.""",
-    ):
-        create_bounding_matrix(5, 10, itakura_max_slope=0.2)
+    expected_result_5_7 = np.array(
+        [
+            [True, False, False, False, False, False, False],
+            [False, True, True, True, True, False, False],
+            [False, False, True, True, True, False, False],
+            [False, False, True, True, True, True, False],
+            [False, False, False, False, False, False, True],
+        ]
+    )
+
+    expected_result_7_5 = np.array(
+        [
+            [True, False, False, False, False],
+            [False, True, False, False, False],
+            [False, True, True, True, False],
+            [False, True, True, True, False],
+            [False, True, True, True, False],
+            [False, False, False, True, False],
+            [False, False, False, False, True],
+        ]
+    )
+
+    matrix = create_bounding_matrix(5, 7, itakura_max_slope=0.5)
+    assert isinstance(matrix, np.ndarray)
+    assert matrix.shape == (5, 7)
+    assert np.array_equal(matrix, expected_result_5_7)
+
+    matrix = create_bounding_matrix(7, 5, itakura_max_slope=0.5)
+    assert isinstance(matrix, np.ndarray)
+    assert matrix.shape == (7, 5)
+    assert np.array_equal(matrix, expected_result_7_5)