def natural_cubic_spline_fit_reg(x, y, dof=5, alpha=0.01):
if not np.any(np.isnan([x, y])):
basis = cr(x, df=dof, constraints="center")
model = Ridge(alpha=alpha).fit(basis, y)
return model.predict(basis)
else:
# Insurance against NAN
# Solution: Fit only non-nan pairs, leave rest as NAN
print("Warning, have NAN")
iNanX = np.isnan(x)
iNanY = np.isnan(y)
iNan = np.logical_or(iNanX, iNanY)
yRez = np.copy(y)
yRez[iNan] = np.nan
yRez[~iNan] = natural_cubic_spline_fit_reg(x[~iNan], y[~iNan], dof=dof, alpha=alpha)
return yRez
### There is a chance that a polygon is repeated twice?
###
X = curr_field['doy']
y = curr_field['EVI']
#############################################
###
### Smoothen
###
#############################################
#
# Spline
#
x_basis = cr(
X, df=freedom_df, constraints='center'
) # Generate spline basis with "freedom_df" degrees of freedom
model = LinearRegression().fit(x_basis, y) # Fit model to the data
spline_pred = model.predict(x_basis) # Get estimates
#
# savitzky
#
# differences are minor, but lets keep using Pythons function
# my_savitzky_pred = rc.savitzky_golay(y, window_size=Sav_win_size, order=sav_order)
savitzky_pred = scipy.signal.savgol_filter(y,
window_length=Sav_win_size,
polyorder=sav_order)
###
### There is a chance that a polygon is repeated twice?
###
X = a_field['doy']
y = a_field['NDVI']
freedom_df = 7
#############################################
###
### Smoothen
###
#############################################
# Generate spline basis with "freedom_df" degrees of freedom
x_basis = cr(X, df=freedom_df, constraints='center')
# Fit model to the data
model = LinearRegression().fit(x_basis, y)
# Get estimates
y_hat = model.predict(x_basis)
#############################################
###
### find peaks
###
#############################################
# peaks_LWLS_1 = peakdetect(LWLS_1[:, 1], lookahead = 10, delta=0)
# max_peaks = peaks_LWLS_1[0]
# peaks_LWLS_1 = form_xs_ys_from_peakdetect(max_peak_list = max_peaks, doy_vect=X)
def powerlaw_random(x, gamma):
return np.power(x, gamma) + 0.0001 * np.random.normal(0, 1, len(x))
plt.close('all')
x = np.linspace(0, 0.5, 100)
y = powerlaw(x, 2 / 3)
y_noise = powerlaw_random(x, 2 / 3)
param, pcov = curve_fit(powerlaw, x, y_noise)
plt.figure()
plt.plot(x, y, 'k')
plt.scatter(x, y_noise, color='r', alpha=0.5)
n_obs = len(x)
df = 10
# Generate spline basis with 10 degrees of freedom
x_basis = cr(x, df=df, constraints='center')
# Fit model to the data
model = LinearRegression().fit(x_basis, y_noise)
# Get estimates
y_hat = model.predict(x_basis)
# plt.scatter(x, y)
plt.plot(x, y_hat, 'r')
plt.title(f'Natural cubic spline with {df} degrees of freedom')
for i in range(0, len(header), 1):
if header[i] != 'DIS':
print(header[i])
xx = data_dictionary[header[i]][:, 0]
# print(x)
index = np.intersect1d(DIS_array[:, 0], xx, return_indices=True)
yy = data_dictionary[header[i]][:, 1] - DIS_array[index[1], 1]
loc = np.where(np.abs(yy) > 1e-10)[0]
x = xx[loc]
y = data_dictionary[header[i]][loc, 1]
df = len(x) - 1
# Generate spline basis with 10 degrees of freedom
x_basis = cr(x, df=df, constraints='center')
# Fit model to the data
model = RANSACRegressor().fit(x_basis, y)
# Get estimates
y_hat = model.predict(x_basis)
# plt.scatter(x, y)
# plt.plot(array_sorted[:,0], y_hat, c[i])
xplot = np.linspace(np.min(x), np.max(x), 1000)
x_basis_large = cr(xplot, df=df, constraints='center')
y_hat = model.predict(x_basis_large)
cs = CubicSpline(x, y)
yplot = cs(xplot)
# # Compare estimated coefficients