You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

191 lines
8.4 KiB
Python

# -*- coding:utf-8 -*-
"""
Created on 5. aug. 2010
@author: pab
"""
import unittest
from numpy.testing import TestCase, assert_array_almost_equal
import wafo.data
import wafo.objects as wo
import wafo.spectrum.models as sm
import wafo.transform.models as tm
class TestTimeSeries(TestCase):
def setUp(self):
x = wafo.data.sea()
self.ts = wo.mat2timeseries(x)
def test_sampling_period(self):
ts = self.ts
assert_array_almost_equal(ts.sampling_period(), 0.25)
def test_tospecdata(self):
S = self.ts.tospecdata(L=150)
print(S.data[:10].tolist())
assert_array_almost_equal(S.data[:10],
[0.0050789888306202345, 0.0049411187454784225,
0.004553923924951667, 0.003990722577978725,
0.00335482379127744, 0.002755110296973988,
0.002281782794825119, 0.0019941282234629933,
0.0019329154962902488, 0.002164040256079313])
# [0.00913087, 0.00881073, 0.00791944,
# 0.00664244, 0.00522429, 0.00389816,
# 0.00282753, 0.00207843, 0.00162678,
# 0.0013916])
def test_tocovdata(self):
rf = self.ts.tocovdata(lag=150)
assert_array_almost_equal(rf.data[:10],
[0.22368637, 0.20838473, 0.17110733,
0.12237803, 0.07024054, 0.02064859,
-0.02218831, -0.0555993, -0.07859847,
-0.09166187])
def test_timeseries_trdata(self):
Hs = 7.0
Sj = sm.Jonswap(Hm0=Hs)
S = Sj.tospecdata() # Make spectrum object from numerical values
S.tr = tm.TrOchi(mean=0, skew=0.16, kurt=0, sigma=Hs/4, ysigma=Hs/4)
xs = S.sim(ns=2**20, iseed=10)
ts = wo.mat2timeseries(xs)
g0, _gemp = ts.trdata(monitor=False) # Not Monitor the development
# Equal weight on all points
g1, _gemp = ts.trdata(method='mnonlinear', gvar=0.5)
# Less weight on the ends
g2, _gemp = ts.trdata(method='nonlinear', gvar=[3.5, 0.5, 3.5])
self.assert_(1.2 < S.tr.dist2gauss() < 1.6)
self.assert_(1.65 < g0.dist2gauss() < 2.05)
self.assert_(0.54 < g1.dist2gauss() < 0.95)
self.assert_(1.5 < g2.dist2gauss() < 1.9)
def test_timeseries_wave_periods(self):
true_t = ([-0.69, -0.86, -1.05],
[0.42, 0.78, 1.37],
[0.09, 0.51, -0.85],
[-0.27, -0.08, 0.32],
[3.84377468, 6.35707656, 4.15490909],
[6.25273295, 6.10295202, 3.36978685],
[2.48364668, 4.74282402, 1.75553431],
[3.76908628, 1.360128, 1.61425254],
[-5.05027968, -9.16405436, -15.60113092],
[7.53392635, 13.90687837, 17.35666522],
[-0.2811934, -7.11392635, -13.12687837],
[4.05027968, 8.47405436, 14.74113092],
[2.03999996, 0.07, 0.05],
[-0.93, -0.07, -0.12],
[1.10999996, 0., -0.07],
[-0.86, -0.02, 0.3],
[0.93, -0.8, -0.2],
[1.10999996, 0., -0.07],
[-0.02, 0.3, -0.34],
[6.10295202, 3.36978685, 3.58501107],
[6.25273295, 6.10295202, 3.36978685],
)
pdefs = ['t2c', 'c2t', 't2t', 'c2c',
'd2d', 'u2u', 'd2u', 'u2d',
'd2t', 't2u', 'u2c', 'c2d',
'm2M', 'M2m', 'm2m', 'M2M', 'all',
]
ts = wo.TimeSeries(self.ts.data[0:400, :2], self.ts.args[:400])
for pdef, truth in zip(pdefs, true_t):
T, _ix = ts.wave_periods(vh=0.0, pdef=pdef)
# print(T[:3,])
assert_array_almost_equal(T[:3], truth)
true_t2 = ([1.10999996, 0., - 0.07],
[-0.02, 0.3, - 0.34],
[6.10295202, 3.369787, 3.585011],
[6.25273295, 6.102952, 3.369787],
[-0.27, -0.08, 0.32],
[-0.27, -0.08, 0.32])
wdefs = ['mw', 'Mw', 'dw', 'uw', 'tw', 'cw', ]
for wdef, truth in zip(wdefs, true_t2):
pdef = '{0}2{0}'.format(wdef[0].lower())
T, _ix = ts.wave_periods(vh=0.0, pdef=pdef, wdef=wdef)
print(T[:3])
assert_array_almost_equal(T[:3], truth)
class TestObjects(TestCase):
def setUp(self):
x = wafo.data.sea()
self.ts = wo.mat2timeseries(x)
def test_cycles_and_levelcrossings(self):
tp = self.ts.turning_points()
assert_array_almost_equal(tp.data[:10],
[-1.200495, 0.839505, -0.090495, -0.020495,
-0.090495, -0.040495, -0.160495, 0.259505,
-0.430495, -0.080495]
)
mm = tp.cycle_pairs()
assert_array_almost_equal(mm.data[:10],
[0.839505, -0.020495, -0.040495, 0.259505,
-0.080495, -0.080495, 0.349505, 0.859505,
0.009505, 0.319505])
true_lcs = (([0., 1., 2., 2., 3., 4., 5., 6., 7., 9.],
[-1.7504945, -1.4404945, -1.4204945, -1.4004945,
-1.3704945, -1.3204945, -1.2704945, -1.2604945,
-1.2504945, -1.2004945]),
([0., 1., 2., 3., 3., 4., 5., 6., 7., 9.],
[-1.7504945, -1.4404945, -1.4204945, -1.4004945,
-1.3704945, -1.3204945, -1.2704945, -1.2604945,
-1.2504945, -1.2004945]),
([1., 2., 3., 4., 4., 5., 6., 7., 9., 11.],
[-1.7504945, -1.4404945, -1.4204945, -1.4004945,
-1.3704945, -1.3204945, -1.2704945, -1.2604945,
-1.2504945, -1.2004945]),
([1., 2., 3., 3., 4., 5., 6., 7., 9., 11.],
[-1.7504945, -1.4404945, -1.4204945, -1.4004945,
-1.3704945, -1.3204945, -1.2704945, -1.2604945,
-1.2504945, -1.2004945]))
for i, true_lc in enumerate(true_lcs):
true_count, true_levels = true_lc
lc = mm.level_crossings(kind=i+1)
assert_array_almost_equal(lc.data[:10], true_count)
assert_array_almost_equal(lc.args[:10], true_levels)
def test_levelcrossings_extrapolate(self):
tp = self.ts.turning_points()
mm = tp.cycle_pairs()
lc = mm.level_crossings()
s = lc.sigma # x[:, 1].std()
ix = slice(0, 1000, 100)
lc_ray = lc.extrapolate(-2 * s, 2 * s, dist='rayleigh')
assert_array_almost_equal(lc_ray.data[ix],
[1.78925398e-37, 9.61028192e-23,
2.05282628e-11, 1.74389448e-03,
5.89169345e+01, 5.240000e+02,
6.72609651e+01, 4.46086175e-01,
2.23463577e-04, 8.45526153e-09])
lc_exp = lc.extrapolate(-2 * s, 2 * s, dist='expon')
lc_gpd = lc.extrapolate(-2 * s, 2 * s, dist='genpareto')
assert_array_almost_equal(lc_exp.data[ix],
[6.51864195e-12, 1.13025876e-08,
1.95974080e-05, 3.39796881e-02,
5.89169345e+01, 5.24000000e+02,
6.43476951e+01, 1.13478843e+00,
2.00122906e-02, 3.52921977e-04])
assert_array_almost_equal(lc_gpd.data[ix],
[0.00000000e+00, 0.00000000e+00,
0.00000000e+00, 0.00000000e+00,
5.89169345e+01, 5.24000000e+02,
6.80484770e+01, 1.41019390e-01,
0.00000000e+00, 0.00000000e+00])
if __name__ == "__main__":
# import sys;sys.argv = ['', 'Test.testName']
unittest.main()