diff --git a/tests/test_evaluator.py b/tests/test_evaluator.py
new file mode 100644
index 0000000000000000000000000000000000000000..5394dc6fa0b36b5f496ac71271381243854e084c
--- /dev/null
+++ b/tests/test_evaluator.py
@@ -0,0 +1,28 @@
+# SPDX-FileCopyrightText: Copyright © 2023 Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+"""Tests for measure functions."""
+
+import numpy
+
+
+def test_centered_maxf1():
+ from ptbench.engine.evaluator import _get_centered_maxf1
+
+ # Multiple max F1
+ f1_scores = numpy.array([0.8, 0.9, 1.0, 1.0, 1.0, 0.3])
+ thresholds = numpy.array([0.2, 0.3, 0.4, 0.5, 0.6, 0.7])
+
+ maxf1, threshold = _get_centered_maxf1(f1_scores, thresholds)
+
+ assert maxf1 == 1.0
+ assert threshold == 0.5
+
+ # Single max F1
+ f1_scores = numpy.array([0.8, 0.9, 1.0, 0.9, 0.7, 0.3])
+ thresholds = numpy.array([0.2, 0.3, 0.4, 0.5, 0.6, 0.7])
+
+ maxf1, threshold = _get_centered_maxf1(f1_scores, thresholds)
+
+ assert maxf1 == 1.0
+ assert threshold == 0.4
diff --git a/tests/test_measures.py b/tests/test_measures.py
deleted file mode 100644
index 69bcdc3c0d6c748f75897fd2b7942497a8fd7f19..0000000000000000000000000000000000000000
--- a/tests/test_measures.py
+++ /dev/null
@@ -1,199 +0,0 @@
-# SPDX-FileCopyrightText: Copyright © 2023 Idiap Research Institute <contact@idiap.ch>
-#
-# SPDX-License-Identifier: GPL-3.0-or-later
-"""Tests for measure functions."""
-
-import random
-import unittest
-
-import numpy
-
-from ptbench.utils.measure import (
- base_measures,
- bayesian_measures,
- beta_credible_region,
- get_centered_maxf1,
-)
-
-
-def test_centered_maxf1():
- # Multiple max F1
- f1_scores = numpy.array([0.8, 0.9, 1.0, 1.0, 1.0, 0.3])
- thresholds = numpy.array([0.2, 0.3, 0.4, 0.5, 0.6, 0.7])
-
- maxf1, threshold = get_centered_maxf1(f1_scores, thresholds)
-
- assert maxf1 == 1.0
- assert threshold == 0.5
-
- # Single max F1
- f1_scores = numpy.array([0.8, 0.9, 1.0, 0.9, 0.7, 0.3])
- thresholds = numpy.array([0.2, 0.3, 0.4, 0.5, 0.6, 0.7])
-
- maxf1, threshold = get_centered_maxf1(f1_scores, thresholds)
-
- assert maxf1 == 1.0
- assert threshold == 0.4
-
-
-class TestFrequentist(unittest.TestCase):
- """Unit test for frequentist base measures."""
-
- def setUp(self):
- self.tp = random.randint(1, 100)
- self.fp = random.randint(1, 100)
- self.tn = random.randint(1, 100)
- self.fn = random.randint(1, 100)
-
- def test_precision(self):
- precision = base_measures(self.tp, self.fp, self.tn, self.fn)[0]
- self.assertEqual((self.tp) / (self.tp + self.fp), precision)
-
- def test_recall(self):
- recall = base_measures(self.tp, self.fp, self.tn, self.fn)[1]
- self.assertEqual((self.tp) / (self.tp + self.fn), recall)
-
- def test_specificity(self):
- specificity = base_measures(self.tp, self.fp, self.tn, self.fn)[2]
- self.assertEqual((self.tn) / (self.tn + self.fp), specificity)
-
- def test_accuracy(self):
- accuracy = base_measures(self.tp, self.fp, self.tn, self.fn)[3]
- self.assertEqual(
- (self.tp + self.tn) / (self.tp + self.tn + self.fp + self.fn),
- accuracy,
- )
-
- def test_jaccard(self):
- jaccard = base_measures(self.tp, self.fp, self.tn, self.fn)[4]
- self.assertEqual(self.tp / (self.tp + self.fp + self.fn), jaccard)
-
- def test_f1(self):
- p, r, s, a, j, f1 = base_measures(self.tp, self.fp, self.tn, self.fn)
- self.assertEqual(
- (2.0 * self.tp) / (2.0 * self.tp + self.fp + self.fn), f1
- )
- self.assertAlmostEqual((2 * p * r) / (p + r), f1) # base definition
-
-
-class TestBayesian:
- """Unit test for bayesian base measures."""
-
- def mean(self, k, lk, lambda_):
- return (k + lambda_) / (k + lk + 2 * lambda_)
-
- def mode1(self, k, lk, lambda_): # (k+lambda_), (l+lambda_) > 1
- return (k + lambda_ - 1) / (k + lk + 2 * lambda_ - 2)
-
- def test_beta_credible_region_base(self):
- k = 40
- lk = 10
- lambda_ = 0.5
- cover = 0.95
- got = beta_credible_region(k, lk, lambda_, cover)
- # mean, mode, lower, upper
- exp = (
- self.mean(k, lk, lambda_),
- self.mode1(k, lk, lambda_),
- 0.6741731038857685,
- 0.8922659692341358,
- )
- assert numpy.isclose(got, exp).all(), f"{got} <> {exp}"
-
- def test_beta_credible_region_small_k(self):
- k = 4
- lk = 1
- lambda_ = 0.5
- cover = 0.95
- got = beta_credible_region(k, lk, lambda_, cover)
- # mean, mode, lower, upper
- exp = (
- self.mean(k, lk, lambda_),
- self.mode1(k, lk, lambda_),
- 0.37137359936800574,
- 0.9774872340008449,
- )
- assert numpy.isclose(got, exp).all(), f"{got} <> {exp}"
-
- def test_beta_credible_region_precision_jeffrey(self):
- # simulation of situation for precision TP == FP == 0, Jeffrey's prior
- k = 0
- lk = 0
- lambda_ = 0.5
- cover = 0.95
- got = beta_credible_region(k, lk, lambda_, cover)
- # mean, mode, lower, upper
- exp = (
- self.mean(k, lk, lambda_),
- 0.0,
- 0.0015413331334360135,
- 0.998458666866564,
- )
- assert numpy.isclose(got, exp).all(), f"{got} <> {exp}"
-
- def test_beta_credible_region_precision_flat(self):
- # simulation of situation for precision TP == FP == 0, flat prior
- k = 0
- lk = 0
- lambda_ = 1.0
- cover = 0.95
- got = beta_credible_region(k, lk, lambda_, cover)
- # mean, mode, lower, upper
- exp = (self.mean(k, lk, lambda_), 0.0, 0.025000000000000022, 0.975)
- assert numpy.isclose(got, exp).all(), f"{got} <> {exp}"
-
- def test_bayesian_measures(self):
- tp = random.randint(100000, 1000000)
- fp = random.randint(100000, 1000000)
- tn = random.randint(100000, 1000000)
- fn = random.randint(100000, 1000000)
-
- _prec, _rec, _spec, _acc, _jac, _f1 = base_measures(tp, fp, tn, fn)
- prec, rec, spec, acc, jac, f1 = bayesian_measures(
- tp, fp, tn, fn, 0.5, 0.95
- )
-
- # Notice that for very large k and l, the base frequentist measures
- # should be approximately the same as the bayesian mean and mode
- # extracted from the beta posterior. We test that here.
- assert numpy.isclose(
- _prec, prec[0]
- ), f"freq: {_prec} <> bays: {prec[0]}"
- assert numpy.isclose(
- _prec, prec[1]
- ), f"freq: {_prec} <> bays: {prec[1]}"
- assert numpy.isclose(_rec, rec[0]), f"freq: {_rec} <> bays: {rec[0]}"
- assert numpy.isclose(_rec, rec[1]), f"freq: {_rec} <> bays: {rec[1]}"
- assert numpy.isclose(
- _spec, spec[0]
- ), f"freq: {_spec} <> bays: {spec[0]}"
- assert numpy.isclose(
- _spec, spec[1]
- ), f"freq: {_spec} <> bays: {spec[1]}"
- assert numpy.isclose(_acc, acc[0]), f"freq: {_acc} <> bays: {acc[0]}"
- assert numpy.isclose(_acc, acc[1]), f"freq: {_acc} <> bays: {acc[1]}"
- assert numpy.isclose(_jac, jac[0]), f"freq: {_jac} <> bays: {jac[0]}"
- assert numpy.isclose(_jac, jac[1]), f"freq: {_jac} <> bays: {jac[1]}"
- assert numpy.isclose(_f1, f1[0]), f"freq: {_f1} <> bays: {f1[0]}"
- assert numpy.isclose(_f1, f1[1]), f"freq: {_f1} <> bays: {f1[1]}"
-
- # We also test that the interval in question includes the mode and the
- # mean in this case.
- assert (prec[2] < prec[1]) and (
- prec[1] < prec[3]
- ), f"precision is out of bounds {_prec[2]} < {_prec[1]} < {_prec[3]}"
- assert (rec[2] < rec[1]) and (
- rec[1] < rec[3]
- ), f"recall is out of bounds {_rec[2]} < {_rec[1]} < {_rec[3]}"
- assert (spec[2] < spec[1]) and (
- spec[1] < spec[3]
- ), f"specif. is out of bounds {_spec[2]} < {_spec[1]} < {_spec[3]}"
- assert (acc[2] < acc[1]) and (
- acc[1] < acc[3]
- ), f"accuracy is out of bounds {_acc[2]} < {_acc[1]} < {_acc[3]}"
- assert (jac[2] < jac[1]) and (
- jac[1] < jac[3]
- ), f"jaccard is out of bounds {_jac[2]} < {_jac[1]} < {_jac[3]}"
- assert (f1[2] < f1[1]) and (
- f1[1] < f1[3]
- ), f"f1-score is out of bounds {_f1[2]} < {_f1[1]} < {_f1[3]}"