test: Improve test coverage (#106)

* test(CostMl, CostCosine): Add test, delete duplicate functions, set random seed

* refactor(pw_constant): inject seed for reproducible random state

* test(test_costs): set random seed

* test(test_costs): running black

* test(costlinear): fix an import mishandled by merge with master

src/ruptures/datasets/pw_constant.py

@@ -1,12 +1,14 @@
 """Piecewise constant signal (with noise)"""
 
 import numpy as np
-from numpy import random as rd
+from numpy.random import RandomState
 
 from ruptures.utils import draw_bkps
 
 
-def pw_constant(n_samples=200, n_features=1, n_bkps=3, noise_std=None, delta=(1, 10)):
+def pw_constant(
+    n_samples=200, n_features=1, n_bkps=3, noise_std=None, delta=(1, 10), seed=None
+):
     """Return a piecewise constant signal and the associated changepoints.
 
     Args:
@@ -15,6 +17,7 @@ def pw_constant(n_samples=200, n_features=1, n_bkps=3, noise_std=None, delta=(1,
         n_bkps (int, optional): number of changepoints
         noise_std (float, optional): noise std. If None, no noise is added
         delta (tuple, optional): (delta_min, delta_max) max and min jump values
+        seed (int): random seed
 
     Returns:
         tuple: signal of shape (n_samples, n_features), list of breakpoints
@@ -27,6 +30,7 @@ def pw_constant(n_samples=200, n_features=1, n_bkps=3, noise_std=None, delta=(1,
     delta_min, delta_max = delta
     # mean value
     center = np.zeros(n_features)
+    rd = RandomState(seed)
     for ind in np.split(tt_, bkps):
         if ind.size > 0:
             # jump value

tests/test_costs.py

@@ -1,101 +1,50 @@
 import pytest
-from ruptures.costs import (
-    CostAR,
-    CostCLinear,
-    CostLinear,
-    CostL1,
-    CostL2,
-    CostLinear,
-    CostNormal,
-    CostRank,
-    CostRbf,
-    cost_factory,
-)
+import numpy as np
+from ruptures.costs import cost_factory, CostLinear
 from ruptures.datasets import pw_constant
 from ruptures.exceptions import NotEnoughPoints
 import numpy as np
 
 
 @pytest.fixture(scope="module")
 def signal_bkps_1D():
-    signal, bkps = pw_constant(n_features=1)
+    signal, bkps = pw_constant(n_features=1, seed=1234567890)
     return signal, bkps
 
 
 @pytest.fixture(scope="module")
 def signal_bkps_1D_noisy():
-    signal, bkps = pw_constant(n_features=1, noise_std=1)
+    signal, bkps = pw_constant(n_features=1, noise_std=1, seed=1234567890)
     return signal, bkps
 
 
 @pytest.fixture(scope="module")
 def signal_bkps_5D():
-    signal, bkps = pw_constant(n_features=5)
+    signal, bkps = pw_constant(n_features=5, seed=1234567890)
     return signal, bkps
 
 
 @pytest.fixture(scope="module")
 def signal_bkps_5D_noisy():
-    signal, bkps = pw_constant(n_features=5, noise_std=1)
+    signal, bkps = pw_constant(n_features=5, noise_std=1, seed=1234567890)
     return signal, bkps
 
 
-cost_classes = {CostAR, CostL1, CostL2, CostNormal, CostRbf, CostRank, CostCLinear}
-cost_names = {"ar", "l1", "l2", "normal", "rbf", "rank", "clinear"}
-
-
-@pytest.mark.parametrize("cost", cost_classes)
-def test_costs_1D(signal_bkps_1D, cost):
-    signal, bkps = signal_bkps_1D
-    cost.fit(signal)
-    cost.fit(signal.flatten())
-    cost.error(0, 100)
-    cost.error(100, signal.shape[0])
-    cost.error(10, 50)
-    cost.sum_of_costs(bkps)
-    with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
-
-
-@pytest.mark.parametrize("cost", cost_classes)
-def test_costs_1D_noisy(signal_bkps_1D_noisy, cost):
-    signal, bkps = signal_bkps_1D_noisy
-    cost.fit(signal)
-    cost.fit(signal.flatten())
-    cost.error(0, 100)
-    cost.error(100, signal.shape[0])
-    cost.error(10, 50)
-    cost.sum_of_costs(bkps)
-    with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
-
-
-@pytest.mark.parametrize("cost", cost_classes)
-def test_costs_5D(signal_bkps_5D, cost):
-    signal, bkps = signal_bkps_5D
-    cost.fit(signal)
-    cost.error(0, 100)
-    cost.error(100, signal.shape[0])
-    cost.error(10, 50)
-    cost.sum_of_costs(bkps)
-    with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
-
-
-@pytest.mark.parametrize("cost", cost_classes)
-def test_costs_5D_noisy(signal_bkps_5D_noisy, cost):
-    signal, bkps = signal_bkps_5D_noisy
-    cost.fit(signal)
-    cost.error(0, 100)
-    cost.error(100, signal.shape[0])
-    cost.error(10, 50)
-    cost.sum_of_costs(bkps)
-    with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
+cost_names = {
+    "ar",
+    "l1",
+    "l2",
+    "normal",
+    "rbf",
+    "rank",
+    "clinear",
+    "mahalanobis",
+    "cosine",
+}
 
 
 @pytest.mark.parametrize("cost_name", cost_names)
-def test_costs_1D(signal_bkps_1D, cost_name):
+def test_costs_1D_names(signal_bkps_1D, cost_name):
     signal, bkps = signal_bkps_1D
     cost = cost_factory(cost_name)
     cost.fit(signal)
@@ -105,11 +54,15 @@ def test_costs_1D(signal_bkps_1D, cost_name):
     cost.error(10, 50)
     cost.sum_of_costs(bkps)
     with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
+        if cost_name == "cosine":
+            cost.min_size = 4
+            cost.error(1, 2)
+        else:
+            cost.error(1, 2)
 
 
 @pytest.mark.parametrize("cost_name", cost_names)
-def test_costs_1D_noisy(signal_bkps_1D_noisy, cost_name):
+def test_costs_1D_noisy_names(signal_bkps_1D_noisy, cost_name):
     signal, bkps = signal_bkps_1D_noisy
     cost = cost_factory(cost_name)
     cost.fit(signal)
@@ -119,11 +72,15 @@ def test_costs_1D_noisy(signal_bkps_1D_noisy, cost_name):
     cost.error(10, 50)
     cost.sum_of_costs(bkps)
     with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
+        if cost_name == "cosine":
+            cost.min_size = 4
+            cost.error(1, 2)
+        else:
+            cost.error(1, 2)
 
 
 @pytest.mark.parametrize("cost_name", cost_names)
-def test_costs_5D(signal_bkps_5D, cost_name):
+def test_costs_5D_names(signal_bkps_5D, cost_name):
     signal, bkps = signal_bkps_5D
     cost = cost_factory(cost_name)
     cost.fit(signal)
@@ -132,11 +89,15 @@ def test_costs_5D(signal_bkps_5D, cost_name):
     cost.error(10, 50)
     cost.sum_of_costs(bkps)
     with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
+        if cost_name == "cosine":
+            cost.min_size = 4
+            cost.error(1, 2)
+        else:
+            cost.error(1, 2)
 
 
 @pytest.mark.parametrize("cost_name", cost_names)
-def test_costs_5D_noisy(signal_bkps_5D_noisy, cost_name):
+def test_costs_5D_noisy_names(signal_bkps_5D_noisy, cost_name):
     signal, bkps = signal_bkps_5D_noisy
     cost = cost_factory(cost_name)
     cost.fit(signal)
@@ -145,7 +106,11 @@ def test_costs_5D_noisy(signal_bkps_5D_noisy, cost_name):
     cost.error(10, 50)
     cost.sum_of_costs(bkps)
     with pytest.raises(NotEnoughPoints):
-        cost.error(1, 2)
+        if cost_name == "cosine":
+            cost.min_size = 4
+            cost.error(1, 2)
+        else:
+            cost.error(1, 2)
 
 
 def test_factory_exception():