def test1(self):
r = [
None,
np.array([1., -0.94200795]),
np.array([1., -1.82697124, 0.93944355]),
np.array([1., -2.7532477, 2.74080825, -0.98598416]),
np.array([1., -3.73744923, 5.47666121, -3.73425407, 0.99819203]),
np.array([
1., -4.73317429, 9.2016863, -9.19738005, 4.72640438,
-0.99752856
]),
np.array([
1., -5.51461918, 12.90426156, -16.40243257, 11.93483034,
-4.70540723, 0.78338097
]),
np.array([
1., -5.39214752, 12.16863091, -14.53657345, 9.37051837,
-2.68798986, -0.07875969, 0.1563373
]),
np.array([
1., -5.4893526, 12.21760094, -12.86527508, 3.54425679,
6.35034463, -7.64479007, 3.5089866, -0.62176513
])
]
data = map(float, open(test_data).read().split())
for order in xrange(1, 9):
assert_array_almost_equal(arburg(data, order), r[order])
def test0(self):
d = np.array([1.00000e-00,
-3.2581e-01,
3.4571e-04,
3.3790e-02,
9.8853e-02])
r = arburg([1,2,3,4,5,6,7,-8], 4)
np.testing.assert_array_almost_equal(r, d, decimal=5)
def test1(self):
r = [None,
np.array([ 1. , -0.94200795]) ,
np.array([ 1. , -1.82697124, 0.93944355]) ,
np.array([ 1. , -2.7532477 , 2.74080825, -0.98598416]) ,
np.array([ 1. , -3.73744923, 5.47666121, -3.73425407, 0.99819203]) ,
np.array([ 1. , -4.73317429, 9.2016863 , -9.19738005, 4.72640438, -0.99752856]) ,
np.array([ 1. , -5.51461918, 12.90426156, -16.40243257, 11.93483034, -4.70540723, 0.78338097]) ,
np.array([ 1. , -5.39214752, 12.16863091, -14.53657345, 9.37051837, -2.68798986, -0.07875969, 0.1563373 ]) ,
np.array([ 1. , -5.4893526 , 12.21760094, -12.86527508, 3.54425679, 6.35034463, -7.64479007, 3.5089866 , -0.62176513])]
data = map(float, open(test_data).read().split())
for order in xrange(1,9):
assert_array_almost_equal(arburg(data, order),r[order])
def fit_baseline(self, x, fs=60., _tsplot=False):
"""
fit_baseline(self, x[, fs=60.][, _tsplot=False])
Finds AR coefficients to build prediction errors and estimates baseline
prediction errors distribution. Returns the baseline entropy.
Parameters
----------
x : array_like
Input data. Should specify steering wheel angle over at least 120 s.
x[:N/2] is used to find the AR coefficients.
x[N/2:] is used to build the baseline prediciton error distribution.
fs : float
sampling rate of x in Hz. Default is 60 Hz. Be sure to specify fs if
x was not sampled at 60 Hz.
Returns
-------
hbas : float
baseline entropy
"""
# unpack relevant parameters
N = len(x)
M = self.M
alpha = self.alpha
resample_fs = self.resample_fs
# apply LP and downsample signal
_x = self._resample(x, fs)
Nx = len(_x)
if len(_x) < resample_fs * 60:
raise Exception('Need at least 60 seconds of steering '
'data to calculate baseline\n(ideally '
'need ~ 120 seconds of data).')
# use first half of x to find coefficients of AR model
# using Burg's method
#
# sign on coeffs returned by arburg is reversed
b_pe = arburg(_x[:Nx / 2], 3)
# use second half of x to calculate baseline entropy
# use AR coefficents to build prediction errors
PE = lfilter(b_pe, 1, _x[Nx / 2:])
# need to build bins to approximate the PE distribution
cdf = percentile(PE) # cdf is a CDF object
pe_alpha = 0.5 * (abs(cdf.find(alpha)) + abs(cdf.find(1. - alpha)))
bin_edges = [-10e12] + list(
np.linspace(-M, M, 2 * M + 1) * pe_alpha) + [10e12]
# now we can calculate the baseline entropy
# np.histogram returns a tuple of histogram counts and bin_edges
Pk = np.histogram(PE, bins=bin_edges)[0] / float(len(PE))
# replace small values to avoid excessively high entropy
Pk[Pk < 1e-3] = 1e-3
# and the baseline entropy is
hbas = np.sum(np.multiply(-Pk, np.log2(Pk)))
# store the relevant information
self.bin_edges = bin_edges
self.b_pe = b_pe
self.cdf = cdf
self.pkbas = Pk
self.hbas = hbas
if _tsplot:
return hbas, self._baseline_tsplot(x[N / 2:], _x[Nx / 2:], PE, fs)
return hbas
def fit_baseline(self, x, fs=60., _tsplot=False):
"""
fit_baseline(self, x[, fs=60.][, _tsplot=False])
Finds AR coefficients to build prediction errors and estimates baseline
prediction errors distribution. Returns the baseline entropy.
Parameters
----------
x : array_like
Input data. Should specify steering wheel angle over at least 120 s.
x[:N/2] is used to find the AR coefficients.
x[N/2:] is used to build the baseline prediciton error distribution.
fs : float
sampling rate of x in Hz. Default is 60 Hz. Be sure to specify fs if
x was not sampled at 60 Hz.
Returns
-------
hbas : float
baseline entropy
"""
# unpack relevant parameters
N = len(x)
M = self.M
alpha = self.alpha
resample_fs = self.resample_fs
# apply LP and downsample signal
_x = self._resample(x, fs)
Nx = len(_x)
if len(_x) < resample_fs * 60:
raise Exception('Need at least 60 seconds of steering '
'data to calculate baseline\n(ideally '
'need ~ 120 seconds of data).')
# use first half of x to find coefficients of AR model
# using Burg's method
#
# sign on coeffs returned by arburg is reversed
b_pe = arburg(_x[:Nx/2], 3)
# use second half of x to calculate baseline entropy
# use AR coefficents to build prediction errors
PE = lfilter(b_pe, 1, _x[Nx/2:])
# need to build bins to approximate the PE distribution
cdf = percentile(PE) # cdf is a CDF object
pe_alpha = 0.5*(abs(cdf.find(alpha)) + abs(cdf.find(1.-alpha)))
bin_edges = [-10e12]+list(np.linspace(-M, M, 2*M+1)*pe_alpha)+[10e12]
# now we can calculate the baseline entropy
# np.histogram returns a tuple of histogram counts and bin_edges
Pk = np.histogram(PE, bins=bin_edges)[0]/float(len(PE))
# replace small values to avoid excessively high entropy
Pk[Pk < 1e-3] = 1e-3
# and the baseline entropy is
hbas = np.sum(np.multiply(-Pk, np.log2(Pk)))
# store the relevant information
self.bin_edges = bin_edges
self.b_pe = b_pe
self.cdf = cdf
self.pkbas = Pk
self.hbas = hbas
if _tsplot:
return hbas, self._baseline_tsplot(x[N/2:], _x[Nx/2:], PE, fs)
return hbas
def test0(self):
d = np.array(
[1.00000e-00, -3.2581e-01, 3.4571e-04, 3.3790e-02, 9.8853e-02])
r = arburg([1, 2, 3, 4, 5, 6, 7, -8], 4)
np.testing.assert_array_almost_equal(r, d, decimal=5)
