def test_default_call_CEEMDAN(): T = np.arange(50) S = np.cos(T*0.1) max_imf = 2 ceemdan = ceemdan(trials=5) ceemdan(S, T, max_imf)
def test_ceemdan_passingArgumentsViaDict(self): trials = 10 noise_kind = 'uniform' spline_kind = 'linear' # Making sure that we are not testing default options ceemdan = ceemdan() self.assertFalse(ceemdan.trials==trials, self.cmp_msg(ceemdan.trials, trials)) self.assertFalse(ceemdan.noise_kind==noise_kind, self.cmp_msg(ceemdan.noise_kind, noise_kind)) self.assertFalse(ceemdan.EMD.spline_kind==spline_kind, self.cmp_msg(ceemdan.EMD.spline_kind, spline_kind)) # Testing for passing attributes via params params = {"trials": trials, "noise_kind": noise_kind, "spline_kind": spline_kind} ceemdan = ceemdan(**params) self.assertTrue(ceemdan.trials==trials, self.cmp_msg(ceemdan.trials, trials)) self.assertTrue(ceemdan.noise_kind==noise_kind, self.cmp_msg(ceemdan.noise_kind, noise_kind)) self.assertTrue(ceemdan.EMD.spline_kind==spline_kind, self.cmp_msg(ceemdan.EMD.spline_kind, spline_kind))
def test_ceemdan_noiseSeed(self): T = np.linspace(0, 1, 100) S = np.sin(2*np.pi*T+ 4**T) + np.cos( (T-0.4)**2) # Compare up to machine epsilon cmpMachEps = lambda x, y: np.abs(x-y)<=2*np.finfo(x.dtype).eps ceemdan = ceemdan(trials=10) # First run random seed cIMF1 = ceemdan(S) # Second run with defined seed, diff than first ceemdan.noise_seed(12345) cIMF2 = ceemdan(S) # Extremly unlikely to have same seed, thus different results msg_false = "Different seeds, expected different outcomes" if cIMF1.shape == cIMF2.shape: self.assertFalse(np.all(cmpMachEps(cIMF1,cIMF2)), msg_false) # Third run with same seed as with 2nd ceemdan.noise_seed(12345) cIMF3 = ceemdan(S) # Using same seeds, thus expecting same results msg_true = "Used same seed, expected same results" self.assertTrue(np.all(cmpMachEps(cIMF2,cIMF3)), msg_true)
def test_ceemdan_completeRun(self): S = np.random.random(200) ceemdan = ceemdan() cIMFs = ceemdan(S) self.assertTrue(cIMFs.shape[0]>1) self.assertTrue(cIMFs.shape[1]==S.size) self.assertTrue('pool' in ceemdan.__dict__)
def test_ceemdan_notParallel(self): S = np.random.random(100) ceemdan = ceemdan(parallel=False) cIMFs = ceemdan(S) self.assertTrue(cIMFs.shape[0]>1) self.assertTrue(cIMFs.shape[1]==S.size) self.assertFalse('pool' in ceemdan.__dict__)
def test_ceemdan_testMaxImf(self): S = np.random.random(100) ceemdan = ceemdan(trials=10) max_imf = 1 cIMFs = ceemdan(S, max_imf=max_imf) self.assertTrue(cIMFs.shape[0]==max_imf+1) max_imf = 3 cIMFs = ceemdan(S, max_imf=max_imf) self.assertTrue(cIMFs.shape[0]==max_imf+1)
def test_ceemdan_origianlSignal(self): T = np.linspace(0, 1, 100) S = 2*np.cos(3*np.pi*T) + np.cos(2*np.pi*T+ 4**T) # Make a copy of S for comparsion Scopy = np.copy(S) # Compare up to machine epsilon cmpMachEps = lambda x, y: np.abs(x-y)<=2*np.finfo(x.dtype).eps ceemdan = ceemdan(trials=10) ceemdan(S) # The original signal should not be changed after the 'ceemdan' function. msg_true = "Expected no change of the original signal" self.assertTrue(np.all(cmpMachEps(Scopy,S)), msg_true)
def test_ceemdan_passingCustomEMD(self): spline_kind = "linear" params = {"spline_kind": spline_kind} ceemdan = ceemdan() self.assertFalse(ceemdan.EMD.spline_kind==spline_kind, "Not"+self.cmp_msg(ceemdan.EMD.spline_kind, spline_kind)) from EMD import emd emd = emd(**params) ceemdan = ceemdan(ext_EMD=emd) self.assertTrue(ceemdan.EMD.spline_kind==spline_kind, self.cmp_msg(ceemdan.EMD.spline_kind, spline_kind))
def test_ceemdan_simpleRun(): T = np.linspace(0, 1, 100) S = np.sin(2*np.pi*T) config = {"processes": 1} ceemdan = ceemdan(trials=10, max_imf=1, **config) ceemdan.EMD.FIXE_H = 5 ceemdan.ceemdan(S)
def test_ceemdan_noiseKind_uniform(): ceemdan = ceemdan() ceemdan.noise_kind = "uniform" ceemdan.generate_noise(1., 100)
def test_ceemdan_constantEpsilon(): S = np.random.random(100) ceemdan = ceemdan(trials=10, max_imf=2) ceemdan.beta_progress = False ceemdan(S)
def test_ceemdan_noiseKind_unknown(self): ceemdan = ceemdan() ceemdan.noise_kind = "bernoulli" with self.assertRaises(ValueError): ceemdan.generate_noise(1., 100)