Merge pull request #40 from felixriese/fix-quantization-error

FIX quantization error calculation

dev-requirements.txt

@@ -3,6 +3,7 @@
 
 # for formatting
 black==22.10.0
+isort==5.12.0
 
 # for the documentation
 sphinx>=4.5.0

pyproject.toml

@@ -21,3 +21,10 @@ exclude = '''
   | .pytest_cache
 )/
 '''
+
+[tool.isort]
+profile = "black"
+
+[tool.pydocstyle]
+convention = "numpy"
+match = 'susi/*'

requirements.txt

@@ -4,7 +4,7 @@ numpy>=1.18.5
 scikit-learn>=0.21.1
 scipy>=1.3.1
 tqdm>=4.45.0
-matplotlib>=3.3.0
+matplotlib==3.7.2
 
 # for the examples
 notebook>=6.0.0

susi/SOMClustering.py

@@ -525,6 +525,7 @@ def _get_node_distance_matrix(
                     np.divide(
                         np.dot(som_array[node[0], node[1]], datapoint),
                         np.multiply(
+                            # TODO check if an axis needs to be set here
                             np.linalg.norm(som_array),
                             np.linalg.norm(datapoint),
                         ),
@@ -927,6 +928,24 @@ def _get_node_neighbors(
         )
         return list(itertools.product(row_range, column_range))
 
+    def _get_weights_per_datapoint(self, datapoints: Sequence) -> list:
+        """Get SOM weights per datapoint.
+
+        Parameters
+        ----------
+        datapoints : array-like matrix, optional (default=True)
+            Samples of shape = [n_samples, n_features].
+
+        Returns
+        -------
+        float
+            Mean quantization error over all datapoints.
+        """
+        return [
+            self.unsuper_som_[bmu[0], bmu[1]]
+            for bmu in self.get_bmus(datapoints)
+        ]
+
     def get_quantization_error(self, X: Optional[Sequence] = None) -> float:
         """Get quantization error for `X` (or the training data).
 
@@ -953,12 +972,10 @@ def get_quantization_error(self, X: Optional[Sequence] = None) -> float:
         if X is None:
             X = self.X_
 
-        weights_per_datapoint = [
-            self.unsuper_som_[bmu[0], bmu[1]] for bmu in self.get_bmus(X)
-        ]
+        weights_per_datapoint = self._get_weights_per_datapoint(X)
 
         quantization_errors = np.linalg.norm(
-            np.subtract(weights_per_datapoint, X)
+            np.subtract(weights_per_datapoint, X), axis=1
         )
 
         return np.mean(quantization_errors)

susi/SOMPlots.py

@@ -2,8 +2,8 @@
 
 from typing import List, Tuple
 
-import matplotlib.pyplot as plt
 import matplotlib
+import matplotlib.pyplot as plt
 import numpy as np
 
 
@@ -47,7 +47,7 @@ def plot_estimation_map(
     for label in cbar.ax.xaxis.get_ticklabels()[::2]:
         label.set_visible(False)
 
-    plt.grid(b=False)
+    plt.grid(visible=False)
 
     return ax
 
@@ -130,7 +130,7 @@ def plot_som_histogram(
     # to be compatible with plt.imshow:
     ax.invert_yaxis()
 
-    plt.grid(b=False)
+    plt.grid(visible=False)
 
     return ax
 

test-requirements.txt

@@ -2,5 +2,6 @@ codecov>=2.1.10
 coverage>=5.3
 flake8==5.0.4
 nbval>=0.9.5
-pytest>=6.0.1
+pytest>=7.2.0
 pytest-cov>=2.10.1
+pytest_mock==3.11.1

tests/test_SOMClustering.py

@@ -4,18 +4,19 @@
 python -m pytest tests/test_SOMClustering.py
 
 """
-import pytest
 import os
 import sys
+
 import numpy as np
+import pytest
 from sklearn.datasets import make_biclusters
 
 sys.path.insert(
     0, os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 )
-import susi
+import susi  # noqa
 
-X, _, _ = make_biclusters((100, 10), 3)
+test_data, _, _ = make_biclusters((100, 10), 3)
 
 
 @pytest.mark.parametrize(
@@ -665,27 +666,60 @@ def test_get_datapoints_from_node(
 )
 def test_get_u_matrix(n_rows, n_columns, mode):
     som = susi.SOMClustering(n_rows=n_rows, n_columns=n_columns)
-    som.fit(X)
+    som.fit(test_data)
     u_matrix = som.get_u_matrix(mode=mode)
     assert isinstance(u_matrix, np.ndarray)
     assert u_matrix.shape == (n_rows * 2 - 1, n_columns * 2 - 1, 1)
 
 
 def test_get_clusters():
     som = susi.SOMClustering()
-    som.fit(X)
-    clusters = som.get_clusters(X)
-    assert len(clusters) == len(X)
+    som.fit(test_data)
+    clusters = som.get_clusters(test_data)
+    assert len(clusters) == len(test_data)
     assert len(clusters[0]) == 2
 
 
-def test_get_quantization_error():
-    # given
-    som = susi.SOMClustering()
-    som.fit(X)
+class TestGetQuantizationError:
+    def test_with_default_data(self) -> None:
+        # given
+        som = susi.SOMClustering()
+        som.fit(test_data)
+
+        # when
+        qerror = som.get_quantization_error()
+
+        # then
+        assert qerror < 0.05
+
+    def test_with_explicit_data(self, mocker) -> None:
+        # given
+        som = susi.SOMClustering()
+        X = np.array(
+            [
+                [8, 9, 7, 8, 6, 3, 0, 4, 1, 6],
+                [3, 2, 1, 2, 5, 7, 0, 0, 3, 3],
+                [0, 7, 5, 1, 4, 5, 7, 5, 8, 5],
+                [9, 1, 1, 7, 5, 9, 8, 9, 3, 3],
+                [2, 7, 7, 2, 3, 3, 3, 3, 3, 3],
+            ]
+        )
+        som.fit(X)
+        weights_per_datapoint = [
+            [8, 5, 0, 4, 5, 7, 0, 2, 0, 2],
+            [4, 0, 7, 4, 8, 0, 4, 2, 2, 8],
+            [5, 3, 1, 2, 7, 9, 9, 8, 8, 7],
+            [8, 6, 8, 7, 7, 5, 4, 7, 1, 1],
+            [6, 2, 3, 5, 7, 5, 9, 5, 9, 0],
+        ]
+        mocker.patch.object(
+            som,
+            "_get_weights_per_datapoint",
+            return_value=weights_per_datapoint,
+        )
 
-    # when
-    qerror = som.get_quantization_error()
+        # when
+        qerror = som.get_quantization_error(X)
 
-    # then
-    assert qerror < 0.05
+        # then
+        assert qerror == 11.45650021348017