from unittest import TestCase,skipIf from unittest.mock import patch import numpy as np from wisp_fast import features as f @skipIf(f._family_kernels is None,'requires optional native backend') class NewKernels(TestCase): def test_spectral_and_center(self): rng=np.random.default_rng(907) for n,c,t in ((1,1,160),(9,3,33),(12,1,128),(17,5,513)): x=rng.normal(size=(n,t,c));x[0,::2]=0. ac=np.empty((n,3,c));center=np.empty_like(x) f._family_kernels.autocorr(x,(1,4,8),ac,center) bands=((0.,.042),(.17,.17),(.5,.51),(.01,.49)) self.assertEqual(center.tobytes(),(x-x.mean(axis=1,keepdims=True)).tobytes()) self.assertEqual(f._spectrum_centered(center,bands).tobytes(),f.ref._spectral_features(x,bands).tobytes()) x=np.zeros((2,31,3)) with self.assertRaises(ValueError):f._family_kernels.autocorr(x,(1,),np.empty((2,1,3)),x) with self.assertRaises(ValueError):f._family_kernels.derived(x,np.empty((2,31,1)),x,np.empty((2,31,1))) def test_window_lane_r(self): with np.errstate(over='ignore',invalid='ignore'): for t,c in ((18,2),(127,5),(160,3),(193,1),(256,3),(257,5)): x=np.random.default_rng(t).normal(size=(17,t,c));x[1]=0.;x[2]*=1.e-130 m=f.RandomConvSketch(n_kernels=17,kernel_lengths=(1,5,7,9,11,15),random_state=14).fit(x) y=m.transform_normalized(x) with patch.object(f,'_cis_conv',None):ref=m.transform_normalized(x) np.testing.assert_array_equal(y,ref) def test_private_standardization(self): from wisp_fast.engine import build_family from sklearn.pipeline import Pipeline from sklearn.preprocessing import StandardScaler from sklearn.linear_model import RidgeClassifier rng=np.random.default_rng(9247);X=rng.normal(size=(20,6));y=np.arange(20)%3 m=build_family('wisp_cis');m.model_=Pipeline([('s',StandardScaler()),('h',RidgeClassifier())]).fit(X,y) m.classes_=np.arange(3);orig=X.copy() a=m._head_proba(X) self.assertEqual(orig.tobytes(),X.tobytes()) b=m._head_proba(X.copy(),private=True);np.testing.assert_array_equal(a,b) X.flags.writeable=False;np.testing.assert_array_equal(a,m._head_proba(X,private=True)) def test_clean_and_noncontiguous(self): a=np.array([[np.nan,np.inf,-0.,-np.inf,1.]]) check=np.nan_to_num(a,nan=0.,posinf=0.,neginf=0.) self.assertEqual(f._clean(a).tobytes(),check.tobytes())