def __kernel_regress(fn, X, y, x):
if fn == 0: kernel = C(1.0, (1e-3, 1e3)) * RBF(1, (1e-2, 1e2))
elif fn == 1: kernel = C(1.0, (1e-3, 1e3)) * RQ(length_scale=5.0, alpha=5.0, length_scale_bounds=(1e-02, 1e2), alpha_bounds=(1e-02, 1e2))
gp = GaussianProcessRegressor(kernel=kernel, n_restarts_optimizer=9)
gp.fit(X, y)
y_pred, sigma = gp.predict(x, return_std=True)
return y_pred[:,0], sigma
def opt_gp(self):
self.nrst = 10
self.kernel = RQ(length_scale=1.0,
alpha=0.1,
length_scale_bounds=(1e-02, 1e2),
alpha_bounds=(1e-2, 1e2))
self.gp = GaussianProcessRegressor(kernel=self.kernel,
n_restarts_optimizer=self.nrst)
return
def get_gpr(kernel_type, hyp, nrst=10, trw=27):
if kernel_type == "RBF":
kernel = RBF(length_scale=hyp["l"], length_scale_bounds=(1e-02, 1e2))
if kernel_type == "RQ":
kernel = RQ(length_scale=hyp["l"],
alpha=hyp["a"],
length_scale_bounds=(1e-02, 1e2),
alpha_bounds=(1e-2, 1e2))
if kernel_type == "Matern":
kernel = Matern(length_scale=hyp["l"],
length_scale_bounds=(1e-02, 1e2),
nu=1.4)
gpr = GaussianProcessRegressor(kernel=kernel, n_restarts_optimizer=nrst)
return (gpr, "GPR", trw)
def ini(self):
self.nrst = 10
if self.k_type == "RBF":
self.kernel = RBF(length_scale=1.0,
length_scale_bounds=(1e-02, 1e2))
if self.k_type == "RQ":
self.kernel = RQ(length_scale=1.0,
alpha=0.1,
length_scale_bounds=(1e-02, 1e2),
alpha_bounds=(1e-2, 1e2))
if self.k_type == "Matern":
self.kernel = Matern(length_scale=1.0,
length_scale_bounds=(1e-02, 1e2),
nu=1.4)
return
def constructModel(self):
optimizedHyperParameters = self.optimizedHyperParameters
fixedHyperParameters = self.fixedHyperParameters
kernelName = optimizedHyperParameters["kernelName"]
if kernelName == 'RBF':
kernel = C(1.0, (1e-5, 1e5)) * RBF(1.0, (1e-5, 1e5))
elif kernelName == 'Matern':
kernel = M(1.0, (1e-5, 1e5), nu=1.5)
elif kernelName == 'RationalQuadratic':
kernel = RQ(1.0, 1.0, (1e-5, 1e5), (1e-5, 1e5))
model = GaussianProcessRegressor(kernel=kernel)
self.model = model
while (n_out > 0 or final_fit==0) and num_finite >= 2:
beta1, beta0, incert_slope, _, _ = ft.wls_matrix(time_vals[good_vals], data_vals[good_vals],
1. / err_vals[good_vals], conf_slope=0.99)
# standardized dispersion from linearity
res_stdized = np.sqrt(np.mean(
(data_vals[good_vals] - (beta0 + beta1 * time_vals[good_vals])) ** 2 / err_vals[good_vals]))
res = np.sqrt(np.mean((data_vals[good_vals] - (beta0 + beta1 * time_vals[good_vals])) ** 2))
if perc_nonlin[min(niter, len(perc_nonlin) - 1)] == 0:
opt_var = 0
else:
opt_var = (res / res_stdized ** 2) ** 2 * 100. / (5 * perc_nonlin[min(niter, len(perc_nonlin) - 1)])
k1 = PairwiseKernel(1, metric='linear') + PairwiseKernel(1, metric='linear') * C(opt_var) * RQ(10, 3) # linear kernel
k2 = C(30) * ESS(length_scale=1, periodicity=1) # periodic kernel
k3 = C(50) * RBF(0.75)
kernel = k1 + k2 + k3
mu_x = np.nanmean(time_vals[good_vals])
detr_t_pred = t_pred - mu_x
detr_time_vals = time_vals - mu_x
mu_y = np.nanmean(data_vals)
detr_data_vals = data_vals - mu_y
# if we remove a linear trend, normalize_y should be false...
gp = GaussianProcessRegressor(kernel=kernel, optimizer=optimizer, n_restarts_optimizer=n_restarts_optimizer,
alpha=err_vals[good_vals], normalize_y=False)
gp.fit(detr_time_vals[good_vals].reshape(-1, 1), detr_data_vals[good_vals].reshape(-1, 1))
y_pred, sigma = gp.predict(detr_t_pred.reshape(-1, 1), return_std=True)
def fit_tile(tile, tmp_ref_dir, base_dir, out_dir):
method = 'gpr'
subspat = None
ref_dem_date = np.datetime64('2013-01-01')
gla_mask = '/calcul/santo/hugonnet/outlines/rgi60_merge.shp'
inc_mask = '/calcul/santo/hugonnet/outlines/rgi60_buff_10.shp'
write_filt = True
clobber = True
tstep = 1. / 12.
time_filt_thresh = [-50, 50]
opt_gpr = False
filt_ref = 'both'
filt_ls = False
conf_filt_ls = 0.99
# specify the exact temporal extent needed to be able to merge neighbouring stacks properly
tlim = [np.datetime64('2000-01-01'), np.datetime64('2020-01-01')]
#for sensitivity test: force final fit only, and change kernel parameters in entry of script
force_final_fit = True
k1 = PairwiseKernel(1, metric='linear') # linear kernel
k2 = C(period_var) * ESS(length_scale=1,
periodicity=1) # periodic kernel
k3 = C(base_var * 0.6) * RBF(base_length * 0.75) + C(
base_var * 0.3) * RBF(base_length * 1.5) + C(base_var * 0.1) * RBF(
base_length * 3)
k4 = PairwiseKernel(1, metric='linear') * C(nonlin_var) * RQ(
nonlin_length, nonlin_alpha)
kernel = k1 + k2 + k3 + k4
lat, lon = SRTMGL1_naming_to_latlon(tile)
epsg, utm = latlon_to_UTM(lat, lon)
print('Fitting tile: ' + tile + ' in UTM zone ' + utm)
# reference DEM
ref_utm_dir = os.path.join(tmp_ref_dir, utm)
ref_vrt = os.path.join(ref_utm_dir, 'tmp_' + utm + '.vrt')
infile = os.path.join(base_dir, utm, tile + '.nc')
outfile = os.path.join(out_dir, utm, tile + '_final.nc')
fn_filt = os.path.join(base_dir, utm, tile + '_filtered.nc')
if True: #not os.path.exists(outfile):
ft.fit_stack(infile,
fn_filt=fn_filt,
fit_extent=subspat,
fn_ref_dem=ref_vrt,
ref_dem_date=ref_dem_date,
exc_mask=gla_mask,
tstep=tstep,
tlim=tlim,
inc_mask=inc_mask,
filt_ref=filt_ref,
time_filt_thresh=time_filt_thresh,
write_filt=True,
outfile=outfile,
method=method,
filt_ls=filt_ls,
conf_filt_ls=conf_filt_ls,
nproc=nproc,
clobber=True,
kernel=kernel,
force_final_fit=force_final_fit)
# write dh/dts for visualisation
ds = xr.open_dataset(outfile)
t0 = np.datetime64('2000-01-01')
t1 = np.datetime64('2020-01-01')
ft.get_full_dh(ds,
os.path.join(
os.path.dirname(outfile),
os.path.splitext(os.path.basename(outfile))[0]),
t0=t0,
t1=t1)
else:
print('Tile already processed.')
if diff - diff_std >0:
base_var = 50 + (diff-diff_std)**2/2
else:
base_var = 50.
else:
base_var = 50.
else:
nonlin_var = (res / res_stdized) ** 2
# nonlin_var = 1
period_nonlinear = 100. / res_stdized ** 2
# period_nonlinear = 10.
base_var=50
print(base_var)
k1 = PairwiseKernel(1, metric='linear') + PairwiseKernel(1, metric='linear') * C(nonlin_var) * RQ(10, period_nonlinear) # linear kernel
k2 = C(30) * ESS(length_scale=1, periodicity=1) # periodic kernel
k3 = C(base_var) * RBF(1.5)
kernel = k1 + k2 + k3
mu_x = np.nanmean(time_vals[good_vals])
detr_t_pred = t_pred - mu_x
detr_time_vals = time_vals - mu_x
mu_y = np.nanmean(data_vals)
detr_data_vals = data_vals - mu_y
# if we remove a linear trend, normalize_y should be false...
gp = GaussianProcessRegressor(kernel=kernel, optimizer=optimizer, n_restarts_optimizer=n_restarts_optimizer,
alpha=err_vals[good_vals], normalize_y=False)
gp.fit(detr_time_vals[good_vals].reshape(-1, 1), detr_data_vals[good_vals].reshape(-1, 1))
y_pred, sigma = gp.predict(detr_t_pred.reshape(-1, 1), return_std=True)
tmp_x = [data[i][j] for j in features]
X.append(tmp_x)
Y0.append(data[i][targets[0]])
Y1.append(data[i][targets[1]])
Y2.append(data[i][targets[2]])
Y3.append(data[i][targets[3]])
# finishing pasing data, now start testing
# Instanciate a Gaussian Process model
#kernel = C(1.0, (1e-3, 1e3)) * RBF(10, (1e-2, 1e2))
#gp = GaussianProcessRegressor(kernel=kernel, n_restarts_optimizer=10)
# TESTIND DIFFERENT KERNELS
printMSE = True
printDetails = False
kernels = [RBF(), MA(), RQ(), RQ(length_scale=50)]
for kernel in kernels:
print kernel
gp = GaussianProcessRegressor(kernel) #n_restarts_optimizer)
clf = gp
print "variable number 1: " + targets[0]
scores = cross_val_score(clf, X, Y0, cv=5, scoring='r2')
if printDetails:
print scores
print "R2: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2)
if printMSE:
scores = cross_val_score(clf,
X,
Y0,
xTrain, yTrain = readTrainFile("Data/problem4b_train.csv")
xTest = readTestFile("Data/problem4b_test.csv")
yTest = readSolutionFile("Data/problem4b_sol.csv")
xTrain = np.atleast_2d(xTrain)
xTest = np.atleast_2d(xTest)
C1 = C(1.0, (1e-3, 1e3))
C2 = C(0.5, (1e-3, 1e3))
RBF1 = RBF(10, (1e-2, 1e2))
RBF2 = RBF(5, (1e-2, 1e2))
RBF3 = RBF(2, (1e-2, 1e2))
RBF4 = RBF(1, (1e-2, 1e2))
RBF5 = RBF(0.5, (1e-2, 1e2))
RQ1 = RQ(10, 0.5, (1e-2, 1e2))
ESS1 = ESS(1.0, 1.0, (1e-05, 100000.0), (1e-05, 100000.0))
kernel1 = C1 * RBF1 + C2
kernel2 = C1 * RBF1 * RBF2 * RBF3 * RBF4 + RBF5
kernel3 = C1 * RQ1 + RBF2
kernel4 = C1 * ESS1 + RBF2
GP = []
for ndx, kernel in zip([1, 2, 3, 4], [kernel1, kernel2, kernel3, kernel4]):
t = time.time()
print(
'---------------------------------------------------------------------------'
)
print(datetime.datetime.now())
print(f'time - {t} :: Fitting GP for kernel - {ndx}')
plt.show() # 数据归一化 data = (data - np.min(data)) / (np.max(data) - np.min(data)) ptr, ttr, pte, tte = split_data(data, n) ptr = np.fliplr(ptr) pte = np.fliplr(pte) best_pop = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1, 1]) p_tr, p_te = translateDNA(best_pop, ptr, pte) # svr clf = svm.SVR() clf.fit(p_tr, ttr) result_svm = clf.predict(p_te) #gpr kernel = RQ(1.0, 1.0, (1e-5, 1e5), (1e-5, 1e5)) reg = GaussianProcessRegressor(kernel=kernel) reg.fit(p_tr, ttr) #这是拟合高斯过程回归的步骤 result_gpr = reg.predict(p_te) #elm Weights, biases, w_out, TF, TYPE = elm.fit(p_tr, ttr) result_elm = elm.predict(p_te, Weights, biases, w_out, TF, TYPE) plt.figure() plt.plot(tte, c='r', label='true') plt.plot(result_svm, c='g', label='svr') plt.plot(result_elm, c='y', label='elm') plt.plot(result_gpr, c='b', label='gpr') plt.legend()
def interp_3d(coord_arr,
val_arr,
resolution=100j,
lattice=None,
xlim=(0, 1),
ylim=(0, 1),
zlim=(0, 1),
method='linear',
return_grids=False):
if lattice is None:
min_x, max_x = xlim
min_y, max_y = ylim
min_z, max_z = zlim
else:
# get the limits of the lattice along the slice direction
x_lat = lattice[:, 0]
y_lat = lattice[:, 1]
z_lat = lattice[:, 2]
# for plotting xlim and ylim
min_x, max_x = min(x_lat), max(x_lat)
min_y, max_y = min(y_lat), max(y_lat)
min_z, max_z = min(z_lat), max(z_lat)
x, y, z = coord_arr[:, 0], coord_arr[:, 1], coord_arr[:, 2]
points = np.array(list(zip(x, y, z)))
# grid on which to evaluate interpolators
grid_x, grid_y, grid_z = np.mgrid[min_x:max_x:resolution * 1j,
min_y:max_y:resolution * 1j,
min_z:max_z:resolution * 1j]
if method == 'kriging':
uk3d = UniversalKriging3D(x,
y,
z,
val_arr,
variogram_model='spherical',
nlags=10,
enable_plotting=False,
drift_terms=['regional_linear'],
verbose=False)
gridx = np.linspace(min_x, max_x, resolution)
gridy = np.linspace(min_y, max_y, resolution)
gridz = np.linspace(min_z, max_z, resolution)
preds, uk_ss3d = uk3d.execute('grid', gridx, gridy, gridz)
preds = preds.data
elif method == 'gp':
kernel = C(1.0, (1e-3, 1e3)) * RQ(2.0, 1.0, (1e-1, 2e1), (1e-3, 1e3))
gp = GaussianProcessRegressor(kernel=kernel,
n_restarts_optimizer=10,
alpha=1e-10)
gp.fit(points, val_arr)
preds = gp.predict(
np.array(list(zip(grid_x.ravel(), grid_y.ravel(),
grid_z.ravel())))).reshape(
resolution, resolution, resolution)
elif method == 'rbf':
rbfi = Rbf(x, y, z, val_arr)
preds = rbfi(grid_x, grid_y, grid_z)
elif method == 'linear':
lndi = LND(points, val_arr)
preds = lndi(grid_x, grid_y, grid_z)
elif method == 'nearest':
nndi = NND(points, val_arr)
preds = nndi(grid_x, grid_y, grid_z)
if return_grids:
return preds, grid_x, grid_y, grid_z
else:
return preds
import matplotlib.pyplot as plt
import pandas as pd
#Scikit Gaussian Process functions
from sklearn.gaussian_process import GaussianProcessRegressor
from sklearn.gaussian_process.kernels import PairwiseKernel, RBF, Product, ConstantKernel as C, RationalQuadratic as RQ, Matern, WhiteKernel
#Scikit MultiOutput Regression class - not needed in the current version
#from sklearn.multioutput import MultiOutputRegressor
sn_lc = pd.read_csv("Users/bhagyasubrayan/Desktop/ZTF/ZTF18abbpeqo.r.txt", sep=" ",
header = 0, names = ['Date','Band','Magnitude','Mag_Error'])
#snlc = sn_lc[['Date','Magnitude','Mag_Error']].to_numpy()
#print(snlc)
x = sn_lc['Date'][:,np.newaxis]
y = sn_lc['Magnitude'][:,np.newaxis]
#print(x)
#print(y)
y_true_stdev = np.std(y)
kern = y_true_stdev*y_true_stdev*RQ(length_scale = 10)*RQ(length_scale = 10)
gp = GaussianProcessRegressor(kernel=kern).fit(x,y)
x_test = (sn_lc['Date'].max() - sn_lc['Date'].min()) * np.random.random_sample((1000,)) + sn_lc['Date'].min()
print(x_test)
y_mean, y_stdev = gp.predict(x_test[:,np.newaxis], return_std = True)
print(y_mean[:,0])
plt.errorbar(x, y, sn_lc['Mag_Error'][:,np.newaxis], fmt='o')
plt.gca().invert_yaxis()
plt.show()
results = np.array([x_test,y_mean[:,0]])
print(results)
plt.errorbar(results[0],results[1],y_stdev, fmt='o')
plt.gca().invert_yaxis()
plt.show()
def constructModel(self):
kernel = RQ(1.0, 1.0, (1e-5, 1e5), (1e-5, 1e5))
self.recurrentModel = GaussianProcessRegressor(kernel=kernel)
res_stdized = np.sqrt(
np.mean(
(data_vals[good_vals] -
(beta0 + beta1 * time_vals[good_vals]))**2 / err_vals[good_vals]))
res = np.sqrt(
np.mean((data_vals[good_vals] -
(beta0 + beta1 * time_vals[good_vals]))**2))
if perc_nonlin[min(niter, len(perc_nonlin) - 1)] == 0:
opt_var = 0
else:
opt_var = (res / res_stdized**2)**2 * 100. / (
5 * perc_nonlin[min(niter,
len(perc_nonlin) - 1)])
k1 = PairwiseKernel(1, metric='linear') + PairwiseKernel(
1, metric='linear') * C(opt_var) * RQ(10, 3) # linear kernel
k2 = C(30) * ESS(length_scale=1, periodicity=1) # periodic kernel
k3 = C(50) * RBF(0.75)
kernel = k1 + k2 + k3
mu_x = np.nanmean(time_vals[good_vals])
detr_t_pred = t_pred - mu_x
detr_time_vals = time_vals - mu_x
mu_y = np.nanmean(data_vals)
detr_data_vals = data_vals - mu_y
# if we remove a linear trend, normalize_y should be false...
gp = GaussianProcessRegressor(kernel=kernel,
optimizer=optimizer,
n_restarts_optimizer=n_restarts_optimizer,
alpha=err_vals[good_vals],