def density(self, j, estimator="linear"):
"""
Compute the signal's density
:param j: scaling parameter [int]
:param estimator: estimator type [str]
:return: density function estimation values [1D array]
"""
[c, d] = self._scaling_coefficients(j)
lowest_value = round(min(self.signal), 3)
highest_value = round(max(self.signal), 3)
density_x = np.linspace(lowest_value, highest_value, self.len_signal)
density_y = np.linspace(lowest_value, highest_value, self.len_signal)
density_z = np.empty(shape=(len(density_x), len(density_y)))
print(
"[INFO] {} - Computing the signal's density using the wavelet {} estimator (resolution level: {})"
.format(functions.get_now(), estimator, j))
if estimator == "donoho" or estimator == "thresholded":
density_z = self._density_donoho(j, c, d, density_x, density_y)
elif estimator == "linear":
density_z = self._density_linear(j, c, density_x, density_y)
else:
print('\n[ERROR] {} - Undefined estimator)'.format(
functions.get_now()))
quit()
density_x, density_y = np.meshgrid(density_x, density_y)
return [density_x, density_y, density_z]
def dwt(self, j):
"""
Compute the Discrete Wavelet Transform
:param j: scaling parameter [int]
:return: approximation coefficients cA, and details coefficients cD [2D array]
"""
print(
'\n[INFO] {} - Computing the Discrete Wavelet Transformation (resolution level: {})'
.format(functions.get_now(), j))
cA = self._dwtA(j)
cD = self._dwtD(j)
return [cA, cD]
def idwt(self, cA, cD, j):
"""
Compute the Inverse Discrete Wavelet Transform
:param cA: approximation coefficients [1D array]
:param cD: details coefficients [1D array]
:param j: scaling parameter [int]
:return: approximation values yA, and details values yD [2D array]
"""
print(
'[INFO] {} - Computing the Inverse Discrete Wavelet Transformation (resolution level: {})'
.format(functions.get_now(), j))
yA = self._idwtA(cA, j)
yD = self._idwtD(cD, j)
return [yA, yD]
def denoise(self, cD, threshold="universal", threshold_type="soft"):
"""
Denoise the signal
:param cD: details coefficients [1D array]
:param threshold: method to compute the threshold value: either "universal" or "SURE" [str]
:param threshold_type: either "soft" or "hard" [str]
:return: denoised details coefficients [1D array]
"""
# Computing the threshold
if threshold == "universal":
print(
"[INFO] {} - Denoising the wavelets coefficients (threshold: universal {})"
.format(functions.get_now(), threshold_type))
threshold_value = self._universal_threshold(cD)
elif threshold == "SURE":
print(
"[INFO] {} - Denoising the wavelets coefficients (threshold: SURE {})"
.format(functions.get_now(), threshold_type))
threshold_value = self._sureshrink_threshold(cD)
else:
threshold_value = 0
print('\n[ERROR] {} - Undefined threshold)'.format(
functions.get_now()))
quit()
# User indication
print('[INFO] {} - Threshold value: {}'.format(functions.get_now(),
threshold_value))
# Applying the threshold
if threshold_type == "soft":
cD = self._soft_thresholding(cD, threshold_value)
elif threshold_type == "hard":
cD = self._hard_thresholding(cD, threshold_value)
else:
print('\n[ERROR] {} - Undefined threshold type)'.format(
functions.get_now()))
quit()
return cD
def _build_father_wavelet(self):
"""
Build the father wavelet
:return: father wavelet matrix [2D array]
"""
# Display indication to user
if self.nb_moments == 1:
nb_iterations = 10
wavelet_name = "haar"
else:
nb_iterations = 6
wavelet_name = "daubechie_{}".format(self.nb_moments)
print('\n[INFO] {} - Building the {} father wavelet ({} iterations)'.
format(functions.get_now(), wavelet_name, nb_iterations))
# Define support
if self.nb_moments == 1:
support = [0, 2]
else:
support = [0, 20]
# Initialization
step = 1
t = np.zeros(support[1] - support[0] + 1)
phi_t = np.zeros(support[1] - support[0] + 1)
for i in range(0, len(t)):
t[i] = support[0] + i
if t[i] == 1:
phi_t[i] = 1
# Cascade algorithm
for j in range(2, nb_iterations + 1):
step = step / 2
t_tampon = t
phi_t_tampon = phi_t
t = np.zeros(2 * len(t_tampon) - 1)
phi_t = np.zeros(2 * len(t_tampon) - 1)
for i in range(0, len(t), 2):
t[i] = t_tampon[int(i / 2)]
phi_t[i] = phi_t_tampon[int(i / 2)]
for i in range(1, len(t) - 1, 2):
t[i] = (t_tampon[int(i / 2)] + t_tampon[int(i / 2) + 1]) / 2
sum_ = 0
for k in range(0, 2 * self.nb_moments):
sum_ = sum_ + coefficients_dict[self.nb_moments][k] * \
self._support_value(2 * t[i] - k, t_tampon, phi_t_tampon)
phi_t[i] = sum_
# Interpolate Daubechie Wavelet to smoothen the function
if self.nb_moments != 1:
for i in range(0, nb_iterations + 1):
for int_index in range(0, support[1]):
if i == 0:
index = self._support_index(int_index, t)
phi_t[index] = (phi_t[index - 1] +
phi_t[index + 1]) / 2
else:
for j in range(1, pow(2, i), 2):
index = self._support_index(
int_index + j / pow(2, i), t)
phi_t[index] = (phi_t[index - 1] +
phi_t[index + 1]) / 2
# Find last used index
check = False
index = len(t) - 1
for i in range(0, len(t)):
if phi_t[i] == 0 and check is False:
index = i
check = True
if phi_t[i] != 0 and check is True:
check = False
# Remove useless indexes
t = t[0:index]
phi_t = phi_t[0:index]
return [t, phi_t]
density_normal_x = np.linspace(min(simulated_returns)*100, max(simulated_returns)*100, len(simulated_returns))
density_normal_y = np.empty(len(density_normal_x))
for i in range(0, len(density_normal_x)):
density_normal_y[i] = math.exp(-(density_normal_x[i] ** 2) / 2) / math.sqrt(2 * math.pi)
density_normal_y[i] = functions.normal_pdf(density_normal_x[i], 0, 1)
density_normal_y = density_normal_y / sum(density_normal_y)
density_normal_x = density_normal_x/100
# Initializing Graph Params
fig, [[ax1, ax2], [ax3, ax4]] = plt.subplots(nrows=2, ncols=2)
fig.set_size_inches(16, 7.6)
fig.set_tight_layout(True)
fig.suptitle(str("Returns Density Estimation"), fontsize=16)
# Simulated Data - Non Parametric Density Estimation (Gaussian Kernel)
print("\n[INFO] {} - Computing the simulated returns density using the gaussian kernel estimator".format(functions.get_now()))
ax1.plot(density_normal_x, density_normal_y, label="Normal Density", linestyle='dashed')
for smoothing in smoothing_array:
[density_kernel_x, density_kernel_y] = functions.compute_kernel_density(simulated_returns, smoothing=smoothing)
functions.initialize_subplot(ax=ax1, title='Simulated Returns - Gaussian Kernel Estimator', xlabel="Returns")
ax1.plot(density_kernel_x, density_kernel_y, label="Smoothing h={}".format(smoothing))
ax1.legend(loc="upper right")
# Simulated Data - Wavelet Parameters
nb_moments = 6
j = 7
# Simulated Data - Wavelet Density Estimations (Wavelet Kernel & Thresholded)
wavelet = Wavelets.Daubechie(signal=simulated_returns, nb_vanishing_moments=nb_moments)
[density_wav_x, density_wav_linear_y] = wavelet.density(j, "linear")
[density_wav_x, density_wav_donoho_y] = wavelet.density(j, "donoho")
functions.initialize_subplot(ax=ax2, title='Simulated Returns - Wavelet Estimators (db_{} / j={})'.format(nb_moments, j), xlabel="Returns")
density_normal_y = np.empty(len(density_normal_x))
for i in range(0, len(density_normal_x)):
density_normal_y[i] = functions.normal_pdf(density_normal_x[i], 0, 1)
density_normal_y = density_normal_y / sum(density_normal_y)
density_normal_x = density_normal_x / 100
# Initializing Graph Params
fig, [[ax1, ax2], [ax3, ax4]] = plt.subplots(nrows=2, ncols=2)
fig.set_size_inches(16, 7.6)
fig.set_tight_layout(True)
fig.suptitle(str("Returns Density Estimation"), fontsize=16)
# Simulated Data - Non Parametric Density Estimation (Gaussian Kernel)
print(
"\n[INFO] {} - Computing the simulated returns density using the gaussian kernel estimator"
.format(functions.get_now()))
ax1.plot(density_normal_x,
density_normal_y,
label="Normal Density",
linestyle='dashed')
for smoothing in smoothing_array:
[density_kernel_x,
density_kernel_y] = functions.compute_kernel_density(simulated_returns,
smoothing=smoothing)
functions.initialize_subplot(
ax=ax1,
title='Simulated Returns - Gaussian Kernel Estimator',
xlabel="Returns")
ax1.plot(density_kernel_x,
density_kernel_y,
label="Smoothing h={}".format(smoothing))
