def cube_stats_f(date_str):
#date_str='20060123 1600'
tim_date=num2date(datestr2num(date_str))
radar1, radar2=netcdf_utis.load_cube('/bm/gdata/scollis/cube_data/'+std_datestr(tim_date, 'uf')+'_winds.nc')
Re=6371.0*1000.0
rad_at_radar=Re*sin(pi/2.0 -abs(radar1['radar_loc'][0]*pi/180.0))#ax_radius(float(lat_cpol), units='degrees')
lons=radar1['radar_loc'][1]+360.0*radar1['xar']/(rad_at_radar*2.0*pi)
lats=radar1['radar_loc'][0] + 360.0*radar1['yar']/(Re*2.0*pi)
angs=array(propigation.make_lobe_grid(radar2['radar_loc'], radar1['radar_loc'], lats,lons))
mywts=met.make_mask_bad1(radar2, radar1, angs, 1.0, 80.0)
submask=met.make_submask(mywts)
X=[radar1['u_array'], radar1['v_array'], radar1['w_array']]
req=[ 'alt(m)', 'wspd(m/s)', 'wdir(degs)', 'tdry(degs)','press(hPa)' ]
first_sonde,second_sonde = read_sounding.get_two_best_conc_sondes(date_str, req_vars=req)
interp_sonde=read_sounding.interp_sonde_time(first_sonde, second_sonde, tim_date, radar1['levs'])
costs=grad_conj_solver_3d.return_cost(X, radar2, radar1, mywts, interp_sonde, submask,loud=True)
disag=costs[1]/(mywts.sum()*2.0)
print disag
def gracon_3d_packaged(X, radar1, radar2, weights, sounding, **kwargs):
#!dimensions
submask = met.make_submask(weights)
nx, ny, nz = radar1["CZ"].shape
Fal = 0.0
i_ter = 1
grad = 1.0
nerr = 0
dimx = 2 * nx * ny * nz
F = 0.0
acc = 0.0
itr = kwargs.get("itr", 100)
#!Initial guess packaging
# do i=1,nx
print "Doing transfer", nx, ny, nz
# X=array([u_array,v_array, w_array])
# gradi=sqrt((G**2).sum())
print "entering loop"
while (i_ter <= itr + 1) and (grad > acc):
if i_ter == 1:
Fm1 = F
#!----------------------COST FUNCTION--------------------
G, F, X = cost_function(X, radar1, radar2, weights, sounding, submask, loud=True)
gradi = sqrt((G ** 2).sum())
#!----------------------COST FUNCTION--------------------
#!subroutine rms2(iterc, f, gradi)
print "Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
c_X = copy(X)
c_F = F
# call rms2(iter,F,gradi)
#! -------- To stop in the first iteration -------------------------
if itr == 0:
grad = -1.0
print "Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
return G, F, X # need to put vars in
#! -----------------------------------------------------------------
# do i=1,dimx
# for i in range(dimx):
XK1 = X
# XK1(i)=X(i)
#! ===============================================================
#! ------------------------Steepest descent-----------------------
if i_ter == 1:
gg = 0.0
# do i=1,dimx
# for i in range(dimx):
gg = gg + (G ** 2).sum() # G[i]*G[i]
D = -G
GK1 = G
gd = -gg
else:
#! ----------------------Conjugate Gradient-----------------------
gg = 0.0
upk = 0.0
dok = 0.0
gg = gg + (G ** 2).sum()
upk = upk + (G * (G - GK1)).sum()
dok = dok + (D * (G - GK1)).sum()
gd = 0.0
# do i=1,dimx
# for i in range(dimx):
GK1 = G
if dok == 0.0:
dok = 1.0e-10
if upk == 0.0:
upk = 1.0e-10
# D(i)=-G(i)+(upk/dok)*D(i)
D = -G + (upk / dok) * D
#!<<<alain
#!dd=dd+D(i)*D(i)
#!>>>alain
gd = gd + (G * D).sum()
#! print*,'Angle between search direction and gradient:',180.0/(4.0*atan(1.0))*acos(-gd/sqrt(dd*gg))
#! --------------------------------------------------------------
# endif
#! ===============================================================
#! ----------------------- step length ---------------------------
if i_ter == 1:
delF = 0.5 * F
elif i_ter != 1:
delF = Fm1 - F
if F != 0.0:
rdF = delF / F
if rdF < 1.0e-5:
print "variation of F too weak"
return G, F, X # need to put vars in
if gd > 0.0:
gd = -gd
axpect = -2.0 * delF / gd
al = min(2.0, axpect)
Fm1 = F
ncomp = 0
subfinish = 0
# do while (subfinish.eq.0)
while subfinish == 0:
# print "in subloop"
ncomp = ncomp + 1
nerr = 0
subfinish = 1
if ncomp > 200:
print "too many subiterations"
print "Current search point: Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
print "Begining of Subloop Cost: ", c_F
if c_F < F:
print "Using start of subloop"
X = c_X
F = c_F
return G, F, X
# endif
# do i=1,dimx
# for i in range(dimx):
# X(i)=XK1(i) + al*D(i)
X = XK1 + al * D
# enddo
# do i=1,nx
G, F, X = cost_function(X, radar1, radar2, weights, sounding, submask)
gradi = sqrt((G ** 2).sum())
#!----------------------COST FUNCTION--------------------
if nerr == 1:
al = al / 2.0
subfinish = 0
# endif
if subfinish == 1:
gald = 0.0
# do i=1,dimx
# for i in range(dimx):
gald = gald + (G * D).sum()
# enddo
if (gald < 0.0) and (Fal < F):
al = 4.0 * al
subfinish = 0
# endif
if subfinish == 1:
z = 3.0 * (F - Fal) / al + gd + gald
w = sqrt(z ** 2 - gd * gald)
amin = al * (1.0 - (gald + w - z) / (gald - gd + 2.0 * w))
# do i=1,dimx
# for i in range(dimx):
# X(i)=XK1(i) + amin*D(i)
X = XK1 + amin * D
# enddo
#!----------------------COST FUNCTION--------------------
#!pack variables
# do i=1,nx
G, F, X = cost_function(X, radar1, radar2, weights, sounding, submask)
gradi = sqrt((G ** 2).sum())
#!----------------------COST FUNCTION--------------------
if (F > Fm1) or (nerr == 1):
al = al / 2.5
subfinish = 0
# endif
if subfinish == 1:
i_ter = i_ter + 1
grad = gg ** 0.5
# endif
# endif
# endif
# enddo
# enddo
# do i=1,nx
print "Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
return G, F, X
def gracon_vel2d_3d(gv_u, gv_v, f, u_array, v_array, i_cmpt_r1, j_cmpt_r1, i_cmpt_r2, j_cmpt_r2, vr1, vr2, weights):
#!dimensions
submask = met.make_submask(weights)
nx, ny, nz = u_array.shape
X = zeros([2 * nx * ny * nz], dtype=float)
G = zeros([2 * nx * ny * nz], dtype=float)
XK1 = zeros([2 * nx * ny * nz], dtype=float)
D = zeros([2 * nx * ny * nz], dtype=float)
GK1 = zeros([2 * nx * ny * nz], dtype=float)
Fal = 0.0
i_ter = 1
grad = 1.0
nerr = 0
dimx = 2 * nx * ny * nz
F = 0.0
acc = 0.0
itr = 100
#!Initial guess packaging
# do i=1,nx
print "Doing transfer", nx, ny, nz
X = array([u_array, v_array])
G = array([u_array, v_array])
gradi = sqrt((G ** 2).sum())
print "entering loop"
while (i_ter <= itr + 1) and (grad > acc):
if i_ter == 1:
Fm1 = F
#!----------------------COST FUNCTION--------------------
#!pack variables
# do i=1,nx
# HERE
u_array = X[0]
v_array = X[1]
gv_u = G[0] * 0.0
gv_v = G[1] * 0.0
F = 0.0
# call meas_cost(gv_u, gv_v, F, u_array, v_array, i_cmpt_r1, j_cmpt_r1,&
# i_cmpt_r2, j_cmpt_r2, nx, ny,nz, vr1, vr2, weights)
gv_u, gv_v, F = meas_cost(
gv_u, gv_v, F, u_array, v_array, i_cmpt_r1, j_cmpt_r1, i_cmpt_r2, j_cmpt_r2, vr1, vr2, weights
)
#!unpack variables
# do i=1,nx
X = array([u_array, v_array])
G = array([gv_u, gv_v])
gradi = sqrt((G ** 2).sum())
#!----------------------COST FUNCTION--------------------
#!subroutine rms2(iterc, f, gradi)
print "Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
# call rms2(iter,F,gradi)
#! -------- To stop in the first iteration -------------------------
if itr == 0:
grad = -1.0
print "Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
return gv_u, gv_v, F, u_array, v_array # need to put vars in
#! -----------------------------------------------------------------
# do i=1,dimx
# for i in range(dimx):
XK1 = X
# XK1(i)=X(i)
#! ===============================================================
#! ------------------------Steepest descent-----------------------
if i_ter == 1:
gg = 0.0
# do i=1,dimx
# for i in range(dimx):
gg = gg + (G ** 2).sum() # G[i]*G[i]
D = -G
GK1 = G
gd = -gg
else:
#! ----------------------Conjugate Gradient-----------------------
gg = 0.0
upk = 0.0
dok = 0.0
gg = gg + (G ** 2).sum()
upk = upk + (G * (G - GK1)).sum()
dok = dok + (D * (G - GK1)).sum()
gd = 0.0
# do i=1,dimx
# for i in range(dimx):
GK1 = G
if dok == 0.0:
dok = 1.0e-10
if upk == 0.0:
upk = 1.0e-10
# D(i)=-G(i)+(upk/dok)*D(i)
D = -G + (upk / dok) * D
#!<<<alain
#!dd=dd+D(i)*D(i)
#!>>>alain
gd = gd + (G * D).sum()
#! print*,'Angle between search direction and gradient:',180.0/(4.0*atan(1.0))*acos(-gd/sqrt(dd*gg))
#! --------------------------------------------------------------
# endif
#! ===============================================================
#! ----------------------- step length ---------------------------
if i_ter == 1:
delF = 0.5 * F
elif i_ter != 1:
delF = Fm1 - F
if F != 0.0:
rdF = delF / F
if rdF < 1.0e-5:
print "variation of F too weak"
return gv_u, gv_v, F, u_array, v_array # need to put vars in
if gd > 0.0:
gd = -gd
axpect = -2 * delF / gd
al = min(2.0, axpect)
Fm1 = F
ncomp = 0
subfinish = 0
# do while (subfinish.eq.0)
while subfinish == 0:
# print "in subloop"
ncomp = ncomp + 1
nerr = 0
subfinish = 1
if ncomp > 200:
print "too many subiterations"
return gv_u, gv_v, F, u_array, v_array # need vars
# endif
# do i=1,dimx
# for i in range(dimx):
# X(i)=XK1(i) + al*D(i)
X = XK1 + al * D
# enddo
# do i=1,nx
u_array = X[0]
v_array = X[1]
gv_u = G[0] * 0.0
gv_v = G[1] * 0.0
F = 0.0
gv_u, gv_v, F = meas_cost(
gv_u, gv_v, F, u_array, v_array, i_cmpt_r1, j_cmpt_r1, i_cmpt_r2, j_cmpt_r2, vr1, vr2, weights
)
X = array([u_array, v_array])
G = array([gv_u, gv_v])
gradi = sqrt((G ** 2).sum())
#!----------------------COST FUNCTION--------------------
if nerr == 1:
al = al / 2
subfinish = 0
# endif
if subfinish == 1:
gald = 0.0
# do i=1,dimx
# for i in range(dimx):
gald = gald + (G * D).sum()
# enddo
if (gald < 0.0) and (Fal < F):
al = 4 * al
subfinish = 0
# endif
if subfinish == 1:
z = 3.0 * (F - Fal) / al + gd + gald
w = sqrt(z ** 2 - gd * gald)
amin = al * (1 - (gald + w - z) / (gald - gd + 2 * w))
# do i=1,dimx
# for i in range(dimx):
# X(i)=XK1(i) + amin*D(i)
X = XK1 + amin * D
# enddo
#!----------------------COST FUNCTION--------------------
#!pack variables
# do i=1,nx
u_array = X[0]
v_array = X[1]
gv_u = G[0] * 0.0
gv_v = G[1] * 0.0
F = 0.0
gv_u, gv_v, F = meas_cost(
gv_u, gv_v, F, u_array, v_array, i_cmpt_r1, j_cmpt_r1, i_cmpt_r2, j_cmpt_r2, vr1, vr2, weights
)
#!unpack variables
# do i=1,nx
X = array([u_array, v_array])
G = array([gv_u, gv_v])
gradi = sqrt((G ** 2).sum())
#!----------------------COST FUNCTION--------------------
if (F > Fm1) or (nerr == 1):
al = al / 2.5
subfinish = 0
# endif
if subfinish == 1:
i_ter = i_ter + 1
grad = gg ** 0.5
# endif
# endif
# endif
# enddo
# enddo
# do i=1,nx
u_array = X[0]
v_array = X[1]
gv_u = G[0]
gv_v = G[1]
print "Iter:", i_ter, " Cost: ", F, " Gradi:", gradi
return gv_u, gv_v, F, u_array, v_array