def drown_2d_field_gridfill(array,
mask=None,
spval=None,
periodic=True,
itermax=100,
relax=.6):
""" fill land values from array, with optional mask, periodicity, ... """
if len(array.shape) != 2:
raise ValueError('array should be 2d')
if mask is not None:
if mask.shape != data.shape:
raise ValueError('mask should be 2d')
if mask.min() < 0 or mask.max() > 1:
raise ValueError('mask values should be 0/1')
else:
mask = logic_mask_from_missing_value(array, spval=spval)
grids = np.ma.masked_array(data=array, mask=mask)
xdim = 1
ydim = 0
eps = 1e-4
drowned, conv = gridfill.fill(grids, xdim, ydim, eps, relax=.6, itermax=100, initzonal=True,\
cyclic=periodic, verbose=False)
return drowned
def test_multi_grid(self):
filled, c = fill(self.grid, 2, 1, self.eps, relax=self.relax,
itermax=self.itermax, initzonal=self.initzonal,
cyclic=self.cyclic, verbose=False)
self.assert_array_almost_equal(filled, self.soln)
def test_not_masked(self):
filled, c = fill(self.grid.filled(fill_value=np.nan), self.eps,
relax=self.relax, itermax=self.itermax,
initzonal=self.initzonal, cyclic=self.cyclic,
verbose=False)
def damp(trend_io, lon, lat):
#Vertical damping from -35.5 to -49.5 lat
#Find latlon indices
ind_start_lat = int(
(np.abs(lat - (-35.5))).argmin()) #start damping from here
ind_end_lat = int(
(np.abs(lat - (-49.5))).argmin()) #last lat with trend data
ind_start_lon = int((np.abs(lon - (20.5))).argmin()) #IO lon1
ind_end_lon = int((np.abs(lon - (146.5))).argmin()) #IO lon2
#Select latlon box to modify
to_modify = trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1]
dmp = np.zeros(to_modify.shape)
c = dmp.shape[0] #number of decrements
for i in range(dmp.shape[1]):
dmp[:, i] = np.linspace(1, 0., c)
modified = np.multiply(to_modify, dmp)
#replace modifed boxes in original dataset
trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1] = modified #dmp_flip
#Horizontal damping from 15.5 to 20.5 longitude
#Find latlon indices
ind_start_lat = int(
(np.abs(lat - (-35.5))).argmin()) #start damping from here
ind_end_lat = int(
(np.abs(lat - (-49.5))).argmin()) #last lat with trend data
ind_start_lon = int((np.abs(lon - (15.5))).argmin()) #IO lon1
ind_end_lon = int((np.abs(lon - (20.5))).argmin()) #IO lon2
#Select latlon box to modify
to_modify = trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1]
dmp = np.zeros(to_modify.shape)
c = dmp.shape[1] #number of decrements
for i in range(dmp.shape[0]):
dmp[i, :] = np.flip(np.linspace(to_modify[i, -1], 0., c), axis=0)
#replace modifed boxes in original dataset
trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1] = dmp #dmp_flip
#Horizontal damping from 141.5 to 146.5 longitude
#Find latlon indices
ind_start_lat = int(
(np.abs(lat - (-35.5))).argmin()) #start damping from here
ind_end_lat = int(
(np.abs(lat - (-49.5))
).argmin()) #last lat with trend data lat with trend data
ind_start_lon = int((np.abs(lon - (146.5))).argmin()) #IO lon1
ind_end_lon = int((np.abs(lon - (151.5))).argmin()) #IO lon2
#Select latlon box to modify
to_modify = trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1]
# #Interpolate masked values
kw = dict(eps=1e-4,
relax=0.6,
itermax=1e4,
initzonal=False,
cyclic=False,
verbose=True)
to_modify_int, converged = fill(np.ma.masked_invalid(to_modify), 1, 0,
**kw)
dmp = np.zeros(to_modify.shape)
c = dmp.shape[1]
for i in range(dmp.shape[0]):
dmp[i, :] = np.linspace(to_modify_int[i, 0], 0., c)
#replace modifed boxes in original dataset
trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1] = dmp #dmp_flip
#Damping near ITF
#Find latlon indices
ind_start_lat = int((np.abs(lat - (20.5))).argmin())
ind_end_lat = int((np.abs(lat - (-20.5))).argmin())
ind_start_lon = int((np.abs(lon - (91.5))).argmin())
ind_end_lon = int((np.abs(lon - (138.5))).argmin())
#Select latlon box to modify
to_modify = trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1]
#Find the coastline
endpoints = np.zeros(to_modify.shape[0], dtype=int)
for i in range(to_modify.shape[0]):
endpoints[i] = np.max(np.where(to_modify[i] > 0))
#Find all end points to damp from
lists = [[] for i in range(len(endpoints))]
for i in range(len(endpoints)):
if i == 0:
lists[i] = [2]
elif i != 0:
if endpoints[i] > endpoints[i - 1]:
lists[i] = (np.arange(endpoints[i - 1], endpoints[i] + 1, 1))
else:
lists[i] = [endpoints[i]]
#damp along diagonals
for i in range(len(lists)):
for j in range(len(lists[i])):
b = i
if i > 7:
a = b - 7
else:
a = 0
c = lists[i][j]
d = c + (b - a)
box = np.nan_to_num(to_modify[a:b + 1, c:d + 1])
box_flip = box[::-1]
dmp = np.linspace(box_flip[0, 0], 0., len(box_flip[0]))
np.fill_diagonal(box_flip, dmp)
box_back = box_flip[::-1]
to_modify[a:b + 1, c:d + 1] = box_back
#mask zeros
to_modify = ma.masked_where(to_modify == 0., to_modify)
#interpolate values
kw = dict(eps=1e-4,
relax=0.6,
itermax=1e4,
initzonal=False,
cyclic=False,
verbose=True)
to_modify_int, converged = fill(np.ma.masked_invalid(to_modify), 1, 0,
**kw)
#replace modified box in original dataset
trend_io[ind_start_lat:ind_end_lat + 1,
ind_start_lon:ind_end_lon + 1] = to_modify_int
return trend_io
def compute(self, pos, coarse_disp, fovea_disp, fovea_ij, disp2imu, imu2disp):
assert all(coarse_disp.shape == self.coarse_shape)
assert all(fovea_disp.shape == self.fovea_shape)
# --- translate fovea_ij from full_shape into fine_shape
fovea_ij = np.round(np.asarray(fovea_ij) * self.fine_full_ratio)
assert all(fovea_ij >= 0)
assert all(fovea_ij + self.fovea_shape < self.fine_shape)
# --- store incoming data
self.data.appendleft((
np.array(pos),
np.array(coarse_disp),
np.array(fovea_disp),
np.array(fovea_ij),
))
# --- project old disparities into new frame
# tic("project")
cur_pos = pos
c_disps_raw = []
c_disps = []
cf_disps = []
cf_foveanesses = []
for k, [pos, coarse, fovea, fovea_ij] in enumerate(self.data):
# --- coarse points
if k == 0: # current frame
c_disp = np.array(coarse)
c_disps_raw.append(np.array(c_disp))
else:
c_xyd = get_shifted_points(
coarse, pos, cur_pos, self.cX, self.cY, disp2imu, imu2disp)
c_xyd[:, :2] = np.round(c_xyd[:, :2])
c_disp = -np.ones(self.coarse_shape)
points2disp_max(c_xyd, c_disp)
c_disps_raw.append(c_disp)
# fill missing values
c_disp_masked = np.ma.masked_array(c_disp, c_disp < 0)
# c_disp, _ = gridfill.fill(mdisp, 1, 0, eps=1, itermax=100)
c_disp, _ = gridfill.fill(c_disp_masked, 1, 0, eps=0.1, itermax=1000)
c_disps.append(c_disp)
# scale up coarse
cf_disp = cv2.resize(c_disp, tuple(self.fine_shape[::-1]))
cf_foveaness = np.zeros_like(cf_disp)
# --- fovea points
if fovea.shape[0] > 0 and fovea.shape[1] > 0:
#import pdb; pdb.set_trace()
fm, fn = fovea.shape
fi, fj = fovea_ij
# account for zero-valued edge around fovea, and edge effects
fmm, fmn = self.fovea_margin
fm = fm - 2*fmm
fn = fn - 2*fmn
fi = fi + fmm
fj = fj + fmn
fovea = fovea[fmm:-fmm, fmn:-fmn]
#import pdb; pdb.set_trace()
if k == 0: # current frame
cf_disp[fi:fi+fm, fj:fj+fn] = fovea
cf_foveaness[fi:fi+fm, fj:fj+fn] = 1
else:
fX, fY = self.fX[fi:fi+fm, fj:fj+fn], self.fY[fi:fi+fm, fj:fj+fn]
f_xyd = get_shifted_points(fovea, pos, cur_pos, fX, fY, disp2imu, imu2disp, final_shape=self.fine_shape)
f_xyd[:, :2] = np.round(f_xyd[:, :2])
f_disp = -np.ones(self.fine_shape)
points2disp_max(f_xyd, f_disp)
cf_disp[f_disp >= 0] = f_disp[f_disp >= 0]
cf_foveaness[f_disp >= 0] = 1
cf_disps.append(cf_disp)
cf_foveanesses.append(cf_foveaness)
# toc()
if 0:
plt.figure(101)
plt.clf()
for k, [c_disp_raw, cf_disp, cf_foveaness] in enumerate(
zip(c_disps, cf_disps, cf_foveanesses)):
r = len(self.data)
c = 3
plt.subplot(r, c, c*k+1)
plt.imshow(c_disp_raw, vmin=-1, vmax=n_disp)
plt.subplot(r, c, c*k+2)
plt.imshow(cf_disp, vmin=-1, vmax=n_disp)
plt.subplot(r, c, c*k+3)
plt.imshow(cf_foveaness, vmin=0, vmax=1)
# --- compute best disparity estimate
# tic("estimate")
self.disp[:] = 0
n = len(cf_disps)
# c_stds = 3 * 2**np.arange(n)
# f_stds = 1 * 2**np.arange(n)
# c_stds = 5 + 5 * np.arange(n)
# f_stds = 0.5 + 5 * np.arange(n)
c_stds = np.array(self.coarse_stds)[:n]
f_stds = np.array(self.fovea_stds)[:n]
# quadratic cost function (estimate is weighted mean of means)
disps = np.array(cf_disps)
foveanesses = np.array(cf_foveanesses)
c_stds.shape = (-1, 1, 1)
f_stds.shape = (-1, 1, 1)
stds = c_stds + foveanesses * (f_stds - c_stds)
w = 1. / stds
w /= w.sum(axis=0, keepdims=True)
self.disp[:] = (w * disps).sum(0)
error = self.disp - disps
# self.cost[:] = (error**2).sum(0) + (stds**2).sum(0)
self.cost[:] = self.disp * ((error**2).sum(0) + (stds**2).sum(0))
# self.cost[:] = (w * (error**2 + stds**2)).sum(0)
# else:
# # compute analytically cost function convolved with each gaussian
# disps = cf_disps
# cs1, cs2 = 1, 0.1 # factors of piecewise quadratic cost function
# costs = np.zeros(self.disp.shape + (n_disp,))
# x = np.arange(n_disp, dtype=float)
# for disp, std in zip(disps, stds):
# # compute three integral points: -inf, x, inf
# a =
# # integral before possible disp
# # integral after possible disp
# # costs +=
# make cost edges zero
fm, fn = (10, 10)
mask = np.ones(self.cost.shape, dtype=bool)
mask[fm:-fm, fn:-fn] = 0
self.cost[mask] = 0
# toc()
return self.disp
#trend_io_3m = xr.full_like(trend_io,3*np.mean(trend_io))
#trend_io_3m = mask_oceans(args.imask, trend_io_3m, lon, lat)
# Damp the edges
trend_io_dmp = damp(trend_io, lon, lat)
#trend_io_dmp = trend_io_dmp.values
#Calculate global climatology
clim = clim_anom_xr(sst_t, time_t, start=None)[0]
clim = clim.values
#clip T
clim[clim < -1.77] = -1.77
#Interpolate sst
clim_int = np.zeros(clim.shape)
for i in range(clim.shape[0]):
clim_int[i], converged = fill(ma.masked_invalid(clim[i]), 1, 0, **kw)
#Repeat the climatology 68 times (1950-2017)
nyears = (end_time - start_time) + 1
clim_rpt = np.repeat(clim_int[np.newaxis,:,:,:],nyears,axis=0)\
.reshape(clim_int.shape[0]*nyears,clim_int.shape[1],clim_int.shape[2])
# Repeat IO trend 68 times (1950-1975)
trend_io_rpt = np.repeat(trend_io_int[np.newaxis, :, :],
nyears * 12,
axis=0)
# Create an array to multiply rates with
multiplier = np.arange(1, (nyears * 12) + 1, 1)[:, np.newaxis, np.newaxis]
# Multiply rate with multiplier to get transient trends
trend_io_trans = np.multiply(trend_io_rpt, multiplier)
def interpolate_to_pressure(data,
presion,
niveles,
extrapolate=False,
decode_times=True):
#interpolar datos a presion
try:
data = xray.open_dataset(data, decode_times=decode_times)
data = data[list(data.data_vars.keys())[0]]
#data=data[var]
except:
print(
"No se encontraron archivos con datos, procediendo a buscar variables"
)
data = data
try:
presion = xray.open_dataset(presion, decode_times=decode_times)
presion = presion[list(presion.data_vars.keys())[0]]
except:
print(
"No se encontraron archivos de presion,procediendo a buscar variables"
)
presion = presion
n = 0
for valor in niveles:
pres_eval_u = presion.where(presion <= valor)
pres_eval_l = presion.where(presion > valor)
temp_u = data.where(pres_eval_u == pres_eval_u.max('lev')).min('lev')
temp_l = data.where(pres_eval_l == pres_eval_l.min('lev')).min('lev')
pres_u = pres_eval_u.max('lev')
pres_l = pres_eval_l.min('lev')
interpolados = (((valor - pres_u) / (pres_l - pres_u)) *
(temp_l - temp_u)) + temp_u
if n == 0:
datos_interp = interpolados.copy()
n += 1
else:
datos_interp = xray.concat([datos_interp, interpolados],
dim='nivel')
shape_cont = len(datos_interp.shape)
datos_interp['nivel'] = niveles
if extrapolate == True:
#extrapolar variables
try:
from gridfill import fill
except ImportError:
raise ImportError(
'Se necesida del modulo gridfill, Repositorio-> https://github.com/ajdawson/gridfill'
)
datos_mascara = np.ma.masked_invalid(datos_interp.values)
kw = dict(eps=1e-4,
relax=0.6,
itermax=1e4,
initzonal=False,
cyclic=False,
verbose=True)
filled, converged = fill(datos_mascara, shape_cont - 1, shape_cont - 2,
**kw)
#datos_interp2=datos_interp.copy()
datos_interp.values = filled
return convertido(datos_interp, data.isel(time=0))
else:
print(datos_interp)
return convertido(datos_interp, data.isel(time=0))
flag = True
except:
new_idx = new_idx - 1
aux_in = data.variables['esi'][dic_esi[newmonth], :, :]
aux_in = ma.masked_where(aux_in <= -6., aux_in)
aux_in = ma.masked_where(aux_in >= 6., aux_in)
#Filling Gap
kw = dict(eps=1e-4,
relax=0.6,
itermax=1e4,
initzonal=False,
cyclic=False,
verbose=True)
aux_in, converged = fill(aux_in, 1, 0, **kw)
aux_in = np.expand_dims(aux_in, axis=0)
aux_in = np.ma.array(aux_in, mask=mask_sa)
aux_in = aux_in[:, min_lat_index:max_lat_index, min_lon_index:max_lon_index]
Dmasked, mask = caso_shape(aux_in[:], lats, lons,
'/nordeste_do_brasil/nordeste_do_brasil.txt')
# Ploting
caso_shape_so_plot('/estados/alagoas/alagoas.txt')
caso_shape_so_plot('/estados/bahia/bahia.txt')
caso_shape_so_plot('/estados/ceara/ceara.txt')
caso_shape_so_plot('/estados/maranhao/maranhao.txt')
caso_shape_so_plot('/estados/paraiba/paraiba.txt')
caso_shape_so_plot('/estados/pernambuco/pernambuco.txt')
caso_shape_so_plot('/estados/piaui/piaui.txt')
new_v = rg_tv(new_vt)
# set options for poisson_grid_fill
guess = True # use zonal means
is_cyclic = True # cyclic [global]
eps = 1.e-2 # variable dependent
opt = 0 # not used
relc = 0.6 # relaxation coefficient
nscan = 1500 # usually much less than this
# fill u v t etc.
ma = np.ma.masked_where(lmask_interp_3d == 0, new_t[0])
filled, converged = gridfill.fill(ma,
2,
1,
eps,
relax=relc,
itermax=nscan,
initzonal=guess,
cyclic=True)
filled = np.where(lmask_out_3d == 0.0, 0.0, filled)
new_t = filled.reshape([1, nz, ny, nx])
ma = np.ma.masked_where(lmask_interp_3d == 0, new_h[0])
filled, converged = gridfill.fill(ma,
2,
1,
eps,
relax=relc,
itermax=nscan,
initzonal=guess,
cyclic=True)
camera = ConnectCam()
views = ["Test"]
Create_Views(views)
from gridfill import fill
kw = dict(eps=1e-4,
relax=0.6,
itermax=1e4,
initzonal=False,
cyclic=False,
verbose=False)
while True:
key = cv2.waitKey(1) & 0xFF
img = cv2.cvtColor(camera.read(), cv2.COLOR_BGR2GRAY)
idx = random_mask(img, frac=0.05)
img = mask_data(img, idx)
img, _ = fill(img, 1, 0, **kw)
print(len(img), len(img[0]))
img = np.array([[int(img[a][i]) for i in range(len(img[0]))]
for a in range(len(img))],
dtype=np.uint8)
print(*img[0])
if key == 'q':
Log_Image("Test", img)
#Update_Views(views,[img])
camera.release()
cv2.destroyAllWindows()
#Start the window thread for the two windows we are using
