From da1c480d08c12302d2fdfce8a837c1e94c739072 Mon Sep 17 00:00:00 2001
From: Adam Taranto <adam.p.taranto@gmail.com>
Date: Sun, 22 Sep 2024 12:54:59 +1000
Subject: [PATCH] Add cosine() tests

---
 src/python/tests/test_metrics.py | 162 +++++++++++++++++++++++++++++++
 1 file changed, 162 insertions(+)
 create mode 100644 src/python/tests/test_metrics.py

diff --git a/src/python/tests/test_metrics.py b/src/python/tests/test_metrics.py
new file mode 100644
index 0000000..abc037e
--- /dev/null
+++ b/src/python/tests/test_metrics.py
@@ -0,0 +1,162 @@
+from math import isclose
+from oxli import KmerCountTable
+from scipy.spatial.distance import cosine
+import numpy as np
+import pytest
+
+
+def test_cosine_similarity_identical_tables():
+    """
+    Test cosine similarity for two identical KmerCountTable objects.
+
+    The cosine similarity should be 1.0 because the vectors representing
+    the k-mer counts are exactly the same, meaning the angle between them
+    is 0 degrees (cos(0) = 1).
+    """
+    kct1 = KmerCountTable(ksize=4)
+    kct2 = KmerCountTable(ksize=4)
+
+    # Manually set k-mer counts
+    kct1["AAAA"] = 5
+    kct1["AATT"] = 3
+    kct1["GGGG"] = 1
+    kct1["CCAA"] = 4
+    kct1["ATTG"] = 6
+
+    # Copy the exact same counts to kct2
+    kct2["AAAA"] = 5
+    kct2["AATT"] = 3
+    kct2["GGGG"] = 1
+    kct2["CCAA"] = 4
+    kct2["ATTG"] = 6
+
+    # Cosine similarity between identical tables should be 1.0
+    # Allow value within 0.001%
+    assert isclose(kct1.cosine(kct2), 1.0, rel_tol=1e-5)
+
+    # Using scipy to calculate the expected value
+    vector1 = [5, 3, 1, 4, 6]
+    vector2 = [5, 3, 1, 4, 6]
+    expected_cosine_sim = 1 - cosine(vector1, vector2)
+
+    # Allow value within 0.001%
+    assert isclose(kct1.cosine(kct2), expected_cosine_sim, rel_tol=1e-5)
+
+
+def test_cosine_similarity_different_tables():
+    """
+    Test cosine similarity for two different KmerCountTable objects.
+
+    The cosine similarity will be less than 1.0 since the vectors
+    are not identical. We will calculate the expected cosine
+    similarity using scipy.
+    """
+    kct1 = KmerCountTable(ksize=4)
+    kct2 = KmerCountTable(ksize=4)
+
+    # Manually set k-mer counts for kct1
+    kct1["AAAA"] = 4
+    kct1["AATT"] = 3
+    kct1["GGGG"] = 1
+    kct1["CCAA"] = 4
+    kct1["ATTG"] = 6
+
+    # Manually set different counts for kct2
+    kct2["AAAA"] = 5
+    kct2["AATT"] = 3
+    kct2["GGGG"] = 1
+    kct2["CCAA"] = 4
+    kct2["ATTG"] = 0
+
+    # Using scipy to calculate the expected value
+    vector1 = [4, 3, 1, 4, 6]
+    vector2 = [5, 3, 1, 4, 0]
+    expected_cosine_sim = 1 - cosine(vector1, vector2)
+
+    # Allow value within 0.001%
+    assert isclose(kct1.cosine(kct2), expected_cosine_sim, rel_tol=1e-5)
+
+
+def test_cosine_similarity_empty_table():
+    """
+    Test cosine similarity for two KmerCountTable objects where one is empty.
+
+    The cosine similarity should be 0.0 because the dot product with an
+    empty table will result in zero, making the numerator of the cosine
+    similarity formula zero.
+    """
+    kct1 = KmerCountTable(ksize=4)
+    kct2 = KmerCountTable(ksize=4)
+
+    # Set counts for kct1
+    kct1["AAAA"] = 5
+    kct1["TTTG"] = 10
+
+    # Leave kct2 empty
+
+    # Cosine similarity should be 0 since one table is empty
+    assert kct1.cosine(kct2) == 0.0
+
+    # Set kct2 with 1 non-overlapping kmer
+    kct2["ATTG"] = 1
+
+    # Cosine similarity should be 0 since no shared kmers
+    assert kct1.cosine(kct2) == 0.0
+
+    # Using scipy for comparison
+    vector1 = [5, 10, 0]
+    vector2 = [0, 0, 1]  # Representing the empty table with no overlap
+    expected_cosine_sim = 1 - cosine(vector1, vector2)
+
+    assert isclose(kct1.cosine(kct2), expected_cosine_sim, rel_tol=1e-5)
+
+
+def test_cosine_similarity_both_empty():
+    """
+    Test cosine similarity for two empty KmerCountTable objects.
+    """
+    # Both tables are empty
+    kct1 = KmerCountTable(ksize=4)
+    kct2 = KmerCountTable(ksize=4)
+
+    # Cosine similarity should be 0.0 for two empty tables
+    assert kct1.cosine(kct2) == 0.0
+
+
+def test_cosine_similarity_partial_overlap():
+    """
+    Test cosine similarity for two KmerCountTable objects with partial overlap in k-mers.
+
+    The cosine similarity should be less than 1.0 but greater than 0.0 because
+    the tables have overlapping k-mers, but their counts differ.
+    """
+    kct1 = KmerCountTable(ksize=4)
+    kct2 = KmerCountTable(ksize=4)
+
+    # Manually set k-mer counts for kct1
+    kct1["AAAA"] = 0  # Not in kct2
+    kct1["AATT"] = 3
+    kct1["GGGG"] = 1
+    kct1["CCAA"] = 4
+    kct1["ATTG"] = 0
+    kct1["AGAT"] = 0  # Set but not in either
+
+    # Manually set k-mer counts for kct2
+    kct2["AAAA"] = 5
+    kct2["AATT"] = 4  # Diff value to kct1
+    kct2["GGGG"] = 1
+    kct2["CCAA"] = 4
+    kct2["ATTG"] = 1  # Not in kct1
+
+    # Using scipy for comparison
+    vector1 = [0, 3, 1, 4, 0, 0]
+    vector2 = [5, 4, 1, 4, 1, 0]
+    expected_cosine_sim = 1 - cosine(vector1, vector2)
+
+    # Cosine similarity is expected to be > 0 but < 1
+    assert isclose(kct1.cosine(kct2), expected_cosine_sim, rel_tol=1e-5)
+
+
+# Jaccard
+
+# Jaccard with setops