Refactor the code basis

.flake8

@@ -0,0 +1,6 @@
+[flake8]
+
+extend-ignore = E221,E211,E222,E202,F541
+
+per-file-ignores =
+    segmetrics/__init__.py:F401

.github/workflows/testsuite.yml

@@ -1,6 +1,5 @@
 name: Test suite
 
-
 on:
 
   workflow_dispatch:
@@ -24,19 +23,39 @@ on:
     branches:
       - master
 
-
 permissions:
   contents: read
   issues: write
   pull-requests: write
 
-
 jobs:
 
+  python_lint:
+
+    name: Linting
+    runs-on: ubuntu-latest
+
+    steps:
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install flake8 isort
+
+      - name: Run flake8
+        shell: bash
+        run: flake8 segmetrics
+
+      - name: Run isort
+        shell: bash
+        run: isort segmetrics --check-only
+
   run_testsuite:
 
     name: Test suite
-
     runs-on: ubuntu-latest
 
     strategy:
@@ -76,4 +95,7 @@ jobs:
           gist-id: f46ddefff0798639bc320b13331dc7ca
           github-auth: ${{ secrets.GITHUB_TOKEN }}
           gist-auth: ${{ secrets.GIST_SECRET }}
-          gist-filename: segmetrics.json
+          gist-filename: segmetrics.json
+          run: |
+            coverage run -m unittest discover
+            python -m coverage json --omit "tests/*,segmetrics/deprecated.py"

.isort.cfg

@@ -0,0 +1,4 @@
+[settings]
+
+multi_line_output = 3
+force_grid_wrap = 2

docs/source/changelog.rst

@@ -72,3 +72,7 @@ Changes default name of ``segmetrics.regional.ISBIScore`` if ``min_ref_size`` is
 
 Changes default name of ``segmetrics.boundary.ObjectBasedDistanceMeasure`` if ``skip_fn`` is not the default.
 
+1.5
+---
+
+Rename ``segmetrics.boundary`` to ``segmetrics.contour``.

docs/source/segmetrics.boundary.rst

@@ -1,7 +1,7 @@
-segmetrics.boundary
-===================
+segmetrics.contour
+==================
 
-.. automodule:: segmetrics.boundary
+.. automodule:: segmetrics.contour
     :members:
     :undoc-members:
     :show-inheritance:

docs/source/segmetrics.rst

@@ -5,6 +5,6 @@ segmetrics API
     segmetrics.study
     segmetrics.measure
     segmetrics.regional
-    segmetrics.boundary
+    segmetrics.contour
     segmetrics.detection
     segmetrics.parallel

docs/source/usage.rst

@@ -54,8 +54,8 @@ Region-based performance measures:
 
 Contour-based performance measures:
 
-- :class:`segmetrics.boundary.Hausdorff`
-- :class:`segmetrics.boundary.NSD`
+- :class:`segmetrics.contour.Hausdorff`
+- :class:`segmetrics.contour.NSD`
 
 Detection-based performance measures:
 
@@ -71,7 +71,7 @@ The choice of suitable performance measaures for evaluation should depend on the
 
 One of the most widely used performance measures is the :py:class:`~segmetrics.regional.Dice` score. This is sensitive to false-positive detections, but invariant to falsely split/merged objects. On the other hand, :py:class:`~segmetrics.regional.ISBIScore` is sensitive to falsely split/merged but invariant to false-positive detections. Thus, using :py:class:`~segmetrics.regional.Dice` in combination with :py:class:`~segmetrics.regional.ISBIScore` well reflects the overall segmentation performance from a region-based point of view.
 
-The :py:class:`~segmetrics.boundary.Hausdorff` distance is very sensitive to outliers (e.g., few objects which yield very high distance values). This high sensitivity is required in some applications (e.g., medical), but it can also cause misleading results in other applications (e.g., cell segmentation). In the latter case, one solution is to use the object-based variant instead (see :ref:`object-based-distance-measures`), which means that such outliers will be averaged out. Another, more simple solution, is to use the quantile-based variant of the :py:class:`~segmetrics.boundary.Hausdorff` distance, which cuts off the outliers based on a carefully chosen quantile value. Suitable choices for the quantile should be between ``0.9`` and ``0.99``, and should be chosen equal for all methods within a comparison. The :py:class:`~segmetrics.boundary.NSD` measure does not suffer from outliers. Using the quantile-based variant of the :py:class:`~segmetrics.boundary.Hausdorff` distance in combination with :py:class:`~segmetrics.boundary.NSD` thus well reflects the overall segmentation performance from a contour-based point of view.
+The :py:class:`~segmetrics.contour.Hausdorff` distance is very sensitive to outliers (e.g., few objects which yield very high distance values). This high sensitivity is required in some applications (e.g., medical), but it can also cause misleading results in other applications (e.g., cell segmentation). In the latter case, one solution is to use the object-based variant instead (see :ref:`object-based-distance-measures`), which means that such outliers will be averaged out. Another, more simple solution, is to use the quantile-based variant of the :py:class:`~segmetrics.contour.Hausdorff` distance, which cuts off the outliers based on a carefully chosen quantile value. Suitable choices for the quantile should be between ``0.9`` and ``0.99``, and should be chosen equal for all methods within a comparison. The :py:class:`~segmetrics.contour.NSD` measure does not suffer from outliers. Using the quantile-based variant of the :py:class:`~segmetrics.contour.Hausdorff` distance in combination with :py:class:`~segmetrics.contour.NSD` thus well reflects the overall segmentation performance from a contour-based point of view.
 
 Including the :py:class:`~segmetrics.detection.FalseSplit` and :py:class:`~segmetrics.detection.FalseMerge` measures is always useful in applications where a main challenge is the separation of the individual objects (e.g., cluster splitting in cell segmentation).
 

requirements.txt

@@ -3,3 +3,4 @@ scipy
 scikit-image>=0.18
 scikit-learn
 dill
+Deprecated==1.2

segmetrics/__init__.py

@@ -1,14 +1,30 @@
-from segmetrics.boundary import Hausdorff, NSD
-from segmetrics.regional import Dice, JaccardCoefficient, JaccardIndex, RandIndex, AdjustedRandIndex, ISBIScore
-from segmetrics.detection import FalsePositive, FalseNegative, FalseSplit, FalseMerge
-from segmetrics.study import Study
-
 import segmetrics.parallel
-
+from segmetrics.contour import (
+    NSD,
+    Hausdorff
+)
+from segmetrics.detection import (
+    FalseMerge,
+    FalseNegative,
+    FalsePositive,
+    FalseSplit
+)
+from segmetrics.regional import (
+    AdjustedRandIndex,
+    Dice,
+    ISBIScore,
+    JaccardCoefficient,
+    JaccardIndex,
+    RandIndex
+)
+from segmetrics.study import Study
 
 VERSION_MAJOR = 1
-VERSION_MINOR = 4
-VERSION_PATCH = 1
-
-VERSION = '%d.%d%s' % (VERSION_MAJOR, VERSION_MINOR, '.%d' % VERSION_PATCH if VERSION_PATCH > 0 else '')
+VERSION_MINOR = 5
+VERSION_PATCH = 0
 
+VERSION = '%d.%d%s' % (
+    VERSION_MAJOR,
+    VERSION_MINOR,
+    '.%d' % VERSION_PATCH if VERSION_PATCH > 0 else '',
+)

segmetrics/_aux.py

@@ -3,8 +3,9 @@
 
 def bbox(*masks, margin=0):
     assert len(masks) > 0
-    assert len(masks) == 1 or all(masks[0].shape == mask.shape for mask in masks[1:])
-    mask_shape = masks[0].shape
+    assert len(masks) == 1 or all(
+        masks[0].shape == mask.shape for mask in masks[1:]
+    )
     _rmin, _rmax = np.inf, -np.inf
     _cmin, _cmax = np.inf, -np.inf
     for mask in masks:
@@ -20,6 +21,5 @@ def bbox(*masks, margin=0):
     _cmax += margin
     _rmin, _rmax = max((_rmin, 0)), min((_rmax, mask.shape[0] - 1))
     _cmin, _cmax = max((_cmin, 0)), min((_cmax, mask.shape[1] - 1))
-    sel = np.s_[_rmin : _rmax + 1, _cmin : _cmax + 1]
+    sel = np.s_[_rmin:_rmax + 1, _cmin:_cmax + 1]
     return sel, (_rmin, _rmax), (_cmin, _cmax)
-