Merge pull request #50 from NorskRegnesentral/fix-scorer-inner-datatype

Fix inner data type in interval scorer and unify cost names

docs/source/api_reference/costs.rst

@@ -11,7 +11,7 @@ Costs
 
     BaseCost
     L2Cost
-    GaussianVarCost
+    GaussianCost
     GaussianCovCost
 
 Utility functions

skchange/anomaly_detectors/anomalisers.py

@@ -63,8 +63,7 @@ def _fit(self, X: pd.DataFrame, y: Optional[pd.DataFrame] = None):
         -------
         self : returns a reference to self
         """
-        self.change_detector_: ChangeDetector
-        self.change_detector_ = self.change_detector.clone()
+        self.change_detector_: BaseChangeDetector = self.change_detector.clone()
         self.change_detector_.fit(X, y)
         return self
 

skchange/anomaly_scores/from_cost.py

@@ -3,7 +3,6 @@
 from typing import Union
 
 import numpy as np
-from numpy.typing import ArrayLike
 
 from skchange.anomaly_scores.base import BaseLocalAnomalyScore, BaseSaving
 from skchange.costs import BaseCost, L2Cost
@@ -81,13 +80,13 @@ def get_param_size(self, p: int) -> int:
         """
         return self.optimised_cost.get_param_size(p)
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the saving scorer.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -218,13 +217,13 @@ def min_size(self) -> int:
         """Minimum valid size of the interval to evaluate."""
         return self.cost.min_size
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the saving scorer.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -339,10 +338,10 @@ def get_test_params(cls, parameter_set="default"):
             `MyClass(**params)` or `MyClass(**params[i])` creates a valid test instance.
             `create_test_instance` uses the first (or only) dictionary in `params`
         """
-        from skchange.costs import GaussianVarCost
+        from skchange.costs import GaussianCost
 
         params = [
             {"cost": L2Cost()},
-            {"cost": GaussianVarCost()},
+            {"cost": GaussianCost()},
         ]
         return params

skchange/anomaly_scores/l2_saving.py

@@ -3,12 +3,10 @@
 __author__ = ["Tveten"]
 
 import numpy as np
-from numpy.typing import ArrayLike
 
 from skchange.anomaly_scores.base import BaseSaving
 from skchange.utils.numba import njit
 from skchange.utils.numba.stats import col_cumsum
-from skchange.utils.validation.data import as_2d_array
 
 
 @njit
@@ -80,13 +78,13 @@ def get_param_size(self, p: int) -> int:
         """
         return p
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the saving evaluator.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -95,8 +93,7 @@ def _fit(self, X: ArrayLike, y=None):
         self :
             Reference to self.
         """
-        X = as_2d_array(X)
-        self.sums_ = col_cumsum(X, init_zero=True)
+        self._sums = col_cumsum(X, init_zero=True)
         return self
 
     def _evaluate(self, cuts: np.ndarray) -> np.ndarray:
@@ -120,7 +117,7 @@ def _evaluate(self, cuts: np.ndarray) -> np.ndarray:
         """
         starts = cuts[:, 0]
         ends = cuts[:, 1]
-        return l2_saving(starts, ends, self.sums_)
+        return l2_saving(starts, ends, self._sums)
 
     @classmethod
     def get_test_params(cls, parameter_set="default"):

skchange/base/base_interval_scorer.py

@@ -72,7 +72,7 @@ def fit(self, X, y=None):
         Parameters
         ----------
         X : pd.Series, pd.DataFrame, or np.ndarray
-            Data to score.
+            Data to evaluate.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -86,22 +86,22 @@ def fit(self, X, y=None):
         Updates the fitted model and sets attributes ending in "_".
         """
         X = check_series(X, allow_index_names=True)
-        self._X = X
+        self._X = as_2d_array(X)
 
-        self._fit(X=X, y=y)
+        self._fit(X=self._X, y=y)
         self._is_fitted = True
         return self
 
-    def _fit(self, X, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the interval scorer to training data.
 
         The core logic of fitting an interval scorer to training data is implemented
         here.
 
         Parameters
         ----------
-        X : pd.Series, pd.DataFrame or np.ndarray
-            Data to score.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 

skchange/base/tests/test_base_interval_evaluator.py

@@ -11,7 +11,7 @@ def _evaluate(self, cuts):
 
 def test_fit():
     evaluator = ConcreteIntervalEvaluator()
-    X = np.array([1, 2, 3, 4, 5])
+    X = np.array([1, 2, 3, 4, 5]).reshape(-1, 1)
     evaluator.fit(X)
     assert evaluator._is_fitted
     assert np.array_equal(evaluator._X, X)

skchange/change_scores/cusum.py

@@ -3,12 +3,10 @@
 __author__ = ["Tveten"]
 
 import numpy as np
-from numpy.typing import ArrayLike
 
 from skchange.change_scores.base import BaseChangeScore
 from skchange.utils.numba import njit
 from skchange.utils.numba.stats import col_cumsum
-from skchange.utils.validation.data import as_2d_array
 
 
 @njit
@@ -74,13 +72,13 @@ def min_size(self) -> int:
         """Minimum size of the interval to evaluate."""
         return 1
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the change score evaluator.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -89,8 +87,7 @@ def _fit(self, X: ArrayLike, y=None):
         self :
             Reference to self.
         """
-        X = as_2d_array(X)
-        self.sums_ = col_cumsum(X, init_zero=True)
+        self._sums = col_cumsum(X, init_zero=True)
         return self
 
     def _evaluate(self, cuts: np.ndarray):
@@ -117,7 +114,7 @@ def _evaluate(self, cuts: np.ndarray):
         starts = cuts[:, 0]
         splits = cuts[:, 1]
         ends = cuts[:, 2]
-        return cusum_score(starts, ends, splits, self.sums_)
+        return cusum_score(starts, ends, splits, self._sums)
 
     @classmethod
     def get_test_params(cls, parameter_set="default"):

skchange/change_scores/from_cost.py

@@ -3,7 +3,6 @@
 from typing import Union
 
 import numpy as np
-from numpy.typing import ArrayLike
 
 from skchange.change_scores.base import BaseChangeScore
 from skchange.costs.base import BaseCost
@@ -57,13 +56,13 @@ def min_size(self) -> int:
         """Minimum valid size of an interval to evaluate."""
         return self.cost.min_size
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the change score.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -124,10 +123,10 @@ def get_test_params(cls, parameter_set="default"):
             `MyClass(**params)` or `MyClass(**params[i])` creates a valid test instance.
             `create_test_instance` uses the first (or only) dictionary in `params`
         """
-        from skchange.costs import GaussianVarCost, L2Cost
+        from skchange.costs import GaussianCost, L2Cost
 
         params = [
             {"cost": L2Cost()},
-            {"cost": GaussianVarCost()},
+            {"cost": GaussianCost()},
         ]
         return params

skchange/change_scores/multivariate_gaussian_score.py

@@ -4,12 +4,10 @@
 __all__ = ["MultivariateGaussianScore"]
 
 import numpy as np
-from numpy.typing import ArrayLike
 
 from skchange.change_scores.base import BaseChangeScore
 from skchange.costs.multivariate_gaussian_cost import MultivariateGaussianCost
 from skchange.utils.numba import njit
-from skchange.utils.validation.data import as_2d_array
 
 
 @njit
@@ -176,13 +174,13 @@ def min_size(self) -> int:
         else:
             return None
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the multivariate Gaussian change score evaluator.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y : None
             Ignored. Included for API consistency by convention.
 
@@ -191,7 +189,7 @@ def _fit(self, X: ArrayLike, y=None):
         self :
             Reference to self.
         """
-        self._cost.fit(as_2d_array(X))
+        self._cost.fit(X)
         return self
 
     def _evaluate(self, cuts: np.ndarray):
@@ -229,7 +227,7 @@ def _evaluate(self, cuts: np.ndarray):
             bartlett_corrections = compute_bartlett_corrections(
                 sequence_lengths=segment_lengths,
                 cut_points=segment_splits,
-                dimension=self._cost.data_dimension_,
+                dimension=self._X.shape[1],
             )
             return bartlett_corrections * raw_scores
         else:

skchange/costs/__init__.py

@@ -1,7 +1,7 @@
 """Cost functions for cost-based change and anomaly detection."""
 
 from skchange.costs.base import BaseCost
-from skchange.costs.gaussian_var_cost import GaussianVarCost
+from skchange.costs.gaussian_cost import GaussianCost
 from skchange.costs.l2_cost import L2Cost
 from skchange.costs.multivariate_gaussian_cost import MultivariateGaussianCost
 
@@ -10,7 +10,7 @@
 ]
 COSTS = [
     MultivariateGaussianCost,
-    GaussianVarCost,
+    GaussianCost,
     L2Cost,
 ]
 

skchange/costs/gaussian_var_cost.py → skchange/costs/gaussian_cost.py

@@ -5,14 +5,12 @@
 from typing import Union
 
 import numpy as np
-from numpy.typing import ArrayLike
 
 from skchange.costs.base import BaseCost
 from skchange.costs.utils import MeanType, VarType, check_mean, check_var
 from skchange.utils.numba import njit
 from skchange.utils.numba.general import truncate_below
 from skchange.utils.numba.stats import col_cumsum
-from skchange.utils.validation.data import as_2d_array
 
 
 @njit
@@ -119,7 +117,7 @@ def gaussian_var_cost_fixed(
     return -log_likelihood
 
 
-class GaussianVarCost(BaseCost):
+class GaussianCost(BaseCost):
     """Univariate Gaussian likelihood cost.
 
     Parameters
@@ -172,23 +170,22 @@ def get_param_size(self, p: int) -> int:
         """
         return 2 * p
 
-    def _fit(self, X: ArrayLike, y=None):
+    def _fit(self, X: np.ndarray, y=None):
         """Fit the cost.
 
         This method precomputes quantities that speed up the cost evaluation.
 
         Parameters
         ----------
-        X : array-like
-            Input data.
+        X : np.ndarray
+            Data to evaluate. Must be a 2D array.
         y: None
             Ignored. Included for API consistency by convention.
         """
-        X = as_2d_array(X)
         self._param = self._check_param(self.param, X)
 
-        self.sums_ = col_cumsum(X, init_zero=True)
-        self.sums2_ = col_cumsum(X**2, init_zero=True)
+        self._sums = col_cumsum(X, init_zero=True)
+        self._sums2 = col_cumsum(X**2, init_zero=True)
         return self
 
     def _evaluate_optim_param(self, starts: np.ndarray, ends: np.ndarray) -> np.ndarray:
@@ -207,7 +204,7 @@ def _evaluate_optim_param(self, starts: np.ndarray, ends: np.ndarray) -> np.ndar
             A 2D array of costs. One row for each interval. The number of
             columns is 1 since the GaussianCovCost is inherently multivariate.
         """
-        return gaussian_var_cost_optim(starts, ends, self.sums_, self.sums2_)
+        return gaussian_var_cost_optim(starts, ends, self._sums, self._sums2)
 
     def _evaluate_fixed_param(self, starts, ends):
         """Evaluate the cost for the fixed parameter.
@@ -226,7 +223,7 @@ def _evaluate_fixed_param(self, starts, ends):
             columns is 1 since the GaussianCovCost is inherently multivariate.
         """
         mean, var = self._param
-        return gaussian_var_cost_fixed(starts, ends, self.sums_, self.sums2_, mean, var)
+        return gaussian_var_cost_fixed(starts, ends, self._sums, self._sums2, mean, var)
 
     @classmethod
     def get_test_params(cls, parameter_set="default"):