Merge pull request #64 from NorskRegnesentral/add-Poisson-cost

Add PoissonCost: Log likelihood of the Poisson distribution, evaluated at the MLE rate

.gitignore

@@ -1,3 +1,6 @@
+# UV Lock file:
+uv.lock
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

skchange/anomaly_scores/tests/test_from_cost.py

@@ -7,11 +7,11 @@
     to_local_anomaly_score,
     to_saving,
 )
-from skchange.costs import COSTS, BaseCost
+from skchange.costs import ALL_COSTS, COSTS, BaseCost
 from skchange.costs.tests.test_all_costs import find_fixed_param_combination
 
 
-@pytest.mark.parametrize("cost_class", COSTS)
+@pytest.mark.parametrize("cost_class", ALL_COSTS)
 def test_saving_init(cost_class):
     param = find_fixed_param_combination(cost_class)
     baseline_cost = cost_class().set_params(**param)
@@ -22,7 +22,7 @@ def test_saving_init(cost_class):
 
 
 @pytest.mark.parametrize("cost_class", COSTS)
-def test_saving_min_size(cost_class):
+def test_saving_min_size(cost_class: type[BaseCost]):
     param = find_fixed_param_combination(cost_class)
     cost = cost_class().set_params(**param)
     saving = Saving(baseline_cost=cost)
@@ -35,7 +35,7 @@ def test_saving_min_size(cost_class):
 
 
 @pytest.mark.parametrize("cost_class", COSTS)
-def test_saving_fit(cost_class):
+def test_saving_fit(cost_class: type[BaseCost]):
     param = find_fixed_param_combination(cost_class)
     cost = cost_class().set_params(**param)
 
@@ -47,7 +47,7 @@ def test_saving_fit(cost_class):
 
 
 @pytest.mark.parametrize("cost_class", COSTS)
-def test_saving_evaluate(cost_class):
+def test_saving_evaluate(cost_class: type[BaseCost]):
     param = find_fixed_param_combination(cost_class)
     cost = cost_class().set_params(**param)
 
@@ -59,8 +59,8 @@ def test_saving_evaluate(cost_class):
     assert savings.shape == (3, 1)
 
 
-@pytest.mark.parametrize("cost_class", COSTS)
-def test_to_saving_raises_with_no_param_set(cost_class):
+@pytest.mark.parametrize("cost_class", ALL_COSTS)
+def test_to_saving_raises_with_no_param_set(cost_class: type[BaseCost]):
     """Test that to_saving raises ValueError with BaseCost that has no param set."""
     with pytest.raises(
         ValueError, match="The baseline cost must have fixed parameters"
@@ -84,17 +84,17 @@ def test_to_saving_error():
         to_saving("invalid_evaluator")
 
 
-@pytest.mark.parametrize("cost_class", COSTS)
-def test_to_local_anomaly_score_with_base_cost(cost_class):
+@pytest.mark.parametrize("cost_class", ALL_COSTS)
+def test_to_local_anomaly_score_with_base_cost(cost_class: type[BaseCost]):
     param = find_fixed_param_combination(cost_class)
     cost = cost_class().set_params(**param)
     local_anomaly_score = to_local_anomaly_score(cost)
     assert isinstance(local_anomaly_score, LocalAnomalyScore)
     assert local_anomaly_score.cost == cost
 
 
-@pytest.mark.parametrize("cost_class", COSTS)
-def test_to_local_anomaly_score_with_local_anomaly_score(cost_class):
+@pytest.mark.parametrize("cost_class", ALL_COSTS)
+def test_to_local_anomaly_score_with_local_anomaly_score(cost_class: type[BaseCost]):
     param = find_fixed_param_combination(cost_class)
     cost = cost_class().set_params(**param)
     local_anomaly_score_instance = LocalAnomalyScore(cost=cost)
@@ -103,7 +103,7 @@ def test_to_local_anomaly_score_with_local_anomaly_score(cost_class):
 
 
 @pytest.mark.parametrize("cost_class", COSTS)
-def test_local_anomaly_score_evaluate(cost_class):
+def test_local_anomaly_score_evaluate(cost_class: type[BaseCost]):
     param = find_fixed_param_combination(cost_class)
     cost = cost_class().set_params(**param)
     local_anomaly_score = LocalAnomalyScore(cost=cost)

skchange/change_scores/tests/test_from_cost.py

@@ -2,7 +2,7 @@
 import pytest
 
 from skchange.change_scores.from_cost import ChangeScore, to_change_score
-from skchange.costs import COSTS
+from skchange.costs import ALL_COSTS, COSTS
 
 
 @pytest.mark.parametrize("cost_class", COSTS)
@@ -16,7 +16,7 @@ def test_change_score_with_costs(cost_class):
     assert scores.shape == (1, 1)
 
 
-@pytest.mark.parametrize("evaluator", COSTS)
+@pytest.mark.parametrize("evaluator", ALL_COSTS)
 def test_to_change_score(evaluator):
     cost_instance = evaluator()
     change_score = to_change_score(cost_instance)

skchange/costs/__init__.py

@@ -7,6 +7,7 @@
 from skchange.costs.laplace_cost import LaplaceCost
 from skchange.costs.multivariate_gaussian_cost import MultivariateGaussianCost
 from skchange.costs.multivariate_t_cost import MultivariateTCost
+from skchange.costs.poisson_cost import PoissonCost
 
 BASE_COSTS = [
     BaseCost,
@@ -19,5 +20,10 @@
     L1Cost,
     L2Cost,
 ]
+INTEGER_COSTS = [
+    PoissonCost,
+]
+
+ALL_COSTS = COSTS + INTEGER_COSTS
 
-__all__ = BASE_COSTS + COSTS
+__all__ = BASE_COSTS + COSTS + INTEGER_COSTS

skchange/costs/laplace_cost.py

@@ -8,7 +8,12 @@
 import numpy as np
 
 from skchange.costs import BaseCost
-from skchange.costs.utils import MeanType, VarType, check_mean, check_univariate_scale
+from skchange.costs.utils import (
+    MeanType,
+    VarType,
+    check_mean,
+    check_non_negative_parameter,
+)
 from skchange.utils.numba import njit
 from skchange.utils.numba.stats import col_median
 from skchange.utils.validation.enums import EvaluationType
@@ -232,7 +237,7 @@ def _check_fixed_param(
         if not isinstance(param, tuple) or len(param) != 2:
             raise ValueError("Fixed Laplace parameters must be (location, scale).")
         means = check_mean(param[0], X)
-        scales = check_univariate_scale(param[1], X)
+        scales = check_non_negative_parameter(param[1], X)
         return means, scales
 
     @property