def coupled_model(forcingtotal, doeclim_out, temperatures, model,
parameters):
if self.aerosol_scaling.value:
forcingtotal = forcing_total.forcing_total(
forcing=forcing,
alpha_doeclim=parameters[7],
l_project=False,
begyear=mod_time[0],
endyear=np.max(mod_time))
if self.climate_sensitivity.value and self.ocean_vertical_diffusivity.value:
doeclim_out = doeclimF.doeclimF(forcingtotal,
self.mod_time,
S=parameters[5],
kappa=parameters[6])
#temperatures = np.array((doeclim_out.loc[(doeclim_out["time"]>=1880) & (doeclim_out["time"]<=2008), "temp"]).tolist())
temperatures = temperatures + parameters[8]
model = gmsl_model.gmsl_model(parameters, temperatures, deltat)
def chain(self, deltat, N=10000):
theta = np.array(self.values)
print('Initial estimate for parameters -', theta)
forcingtotal = forcing_total.forcing_total(
forcing=self.forcing,
alpha_doeclim=self.values[7],
l_project=False,
begyear=self.mod_time[0],
endyear=np.max(self.mod_time))
doeclim_out = doeclimF.doeclimF(forcingtotal,
self.mod_time,
S=self.values[5],
kappa=self.values[6])
temp_out = np.array(
(doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008), "temp"]).tolist())
ocheat = np.array((doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008),
'ocheat.mixed']).tolist())
ocheat = ocheat[self.offset[0]:self.offset[0] + 44]
temp_out += self.T_0
gmsl_out = gmsl_model.gmsl_model(theta, temp_out, 1)
lp = self.logp(theta, deltat, temp_out, gmsl_out, ocheat)
theta_best = theta
lp_max = lp
theta_new = [0.] * len(theta)
accepts = 0
mcmc_chains = np.zeros((N, (len(theta))))
step = np.diag(self.stepsizes)
sd = 2.38**2 / len(theta)
#step = 0.5 * step
#Check if converged. If not keep running.
print(N)
for i in (range(N)):
#print(i)
if i > 500: step = self.update_cov(mcmc_chains[:i], sd, len(theta))
theta_new = list(np.random.multivariate_normal(theta, step))
doeclim_out = doeclimF.doeclimF(forcingtotal,
self.mod_time,
S=theta[5],
kappa=theta[6])
temp_out = np.array(
(doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008),
"temp"]).tolist())
temp_out += theta[8]
ocheat = np.array((doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008),
'ocheat.mixed']).tolist())
gmsl_out = gmsl_model.gmsl_model(theta, temp_out, 1)
ocheat = ocheat[self.offset[0]:self.offset[0] + 44]
lp_new = self.logp(theta_new, deltat, temp_out, gmsl_out, ocheat)
if np.isinf(lp_new):
mcmc_chains[i, :] = theta
continue
#print(lp, lp_new, 'difference')
lq = np.abs(lp_new - lp)
lr = np.log(np.random.uniform(1, 2))
#print(lr, lq)
if (lr < lq):
theta = theta_new
lp = lp_new
accepts += 1
#print('NEW VALUE')
if lp > lp_max:
theta_best = theta
lp_max = lp
mcmc_chains[i, :] = theta
return mcmc_chains, accepts / N * 100
def chain(parameters,
temperatures,
deltat,
sealevel,
sealevel_sigma,
pamnames,
N=10000):
if 'aerosol_scaling' in pamnames:
forcingtotal = forcing_total.forcing_total(
forcing=forcing,
alpha_doeclim=aerosol_scaling,
l_project=False,
begyear=mod_time[0],
endyear=np.max(mod_time))
if 'climate_sensitivity' in pamnames and 'ocean_vertical_diffusivity' in pamnames:
doeclim_out = doeclimF.doeclimF(forcingtotal,
mod_time,
S=climate_sensitivity,
kappa=ocean_vertical_diffusivity)
temperatures = np.array(
(doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008), "temp"]).tolist())
temperatures = temperatures + T_0
theta = parameters
print('Initial estimate for parameters -', theta)
model = gmsl_model.gmsl_model(theta, temperatures,
deltat) if 'alpha' in pamnames else []
lp = logp(theta,
sealevel,
deltat,
temperatures,
model,
pamnames,
sigma=sealevel_sigma)
theta_best = theta
lp_max = lp
theta_new = [0.] * len(parameters)
accepts = 0
mcmc_chains = np.array([np.zeros(len(parameters))] * N)
stepsizes = {
'alpha': .01,
'Teq': .05,
'S0': 5,
'rho': .001,
'sigma_ar': .1,
'climate_sensitivity': 0.16,
'ocean_vertical_diffusivity': 0.17,
'aerosol_scaling': 0.025,
'T_0': 0.03,
'sigma_T': 5e-4,
'rho_T': 0.007
}
step = []
count = 0
for i in pamnames:
temp = [0] * len(pamnames)
temp[count] = stepsizes[i]
count += 1
step.append(temp)
step = np.array(step)
#step = np.array([[.01,0,0,0,0,0,0,0,0], [0,.05,0,0,0,0,0,0,0], [0,0,5,0,0,0,0,0,0], [0,0,0,0.001,0,0,0,0,0],[0,0,0,0,0.1,0,0,0,0],
# [0,0,0,0,0,0.16,0,0,0], [0,0,0,0,0,0,0.17,0,0], [0,0,0,0,0,0,0,0.025,0],[0,0,0,0,0,0,0,0,0.03]])
sd = 2.38**2 / len(theta)
#Check if converged. If not keep running.
print(N)
print(theta)
for i in tqdm(range(N)):
if i > 500: step = update_cov(mcmc_chains[:i], sd, len(parameters))
theta_new = list(np.random.multivariate_normal(theta, step))
if len(pamnames) > 10:
coupled_model(forcingtotal, doeclim_out, temperatures, model,
parameters, mod_time, T_0, pamnames)
model = gmsl_model.gmsl_model(theta, temperatures,
deltat) if 'alpha' in pamnames else []
#Separate above 3 lines + T_0 into a new file.
lp_new = logp(theta_new,
sealevel,
deltat,
temperatures,
model,
pamnames,
sigma=sealevel_sigma)
lq = lp_new - lp
lr = np.math.log(np.random.uniform(0, 1))
#print(lr, lq)
if (lr < lq):
theta = theta_new
lp = lp_new
accepts += 1
if lp > lp_max:
theta_best = theta
lp_max = lp
mcmc_chains[i] = theta
return mcmc_chains, accepts / N * 100
def chain(self, deltat, N=10000):
if self.aerosol_scaling.value:
forcingtotal = forcing_total.forcing_total(
forcing=forcing,
alpha_doeclim=self.aerosol_scaling,
l_project=False,
begyear=mod_time[0],
endyear=np.max(mod_time))
if self.climate_sensitivity.value and self.ocean_vertical_diffusivity.value:
doeclim_out = doeclimF.doeclimF(forcingtotal,
self.mod_time,
S=self.climate_sensitivity.value,
kappa=ocean_vertical_diffusivity)
temperatures = self.temperatures + self.T_0.value
parameters = self.values.values()
parameters = list(filter(lambda a: a != 0, parameters))
self.num = len(parameters)
theta = parameters
print('Initial estimate for parameters -', theta)
model = gmsl_model.gmsl_model(theta, temperatures,
deltat) if self.alpha.value else []
lp = self.logp(theta, deltat, temperatures, model)
theta_best = theta
lp_max = lp
theta_new = [0.] * len(parameters)
accepts = 0
mcmc_chains = np.array([np.zeros(len(parameters))] * N)
step = []
count = 0
temp = self.stepsizes.values()
for i in range(len(parameters)):
temp = [0] * (len(parameters))
temp[count] = temp[i]
count += 1
step.append(temp)
step = np.array(step)
sd = 2.38**2 / len(theta)
#Check if converged. If not keep running.
print(N)
print(theta)
for i in tqdm(range(N)):
if i > 500:
step = self.update_cov(mcmc_chains[:i], sd, len(parameters))
theta_new = list(np.random.multivariate_normal(theta, step))
if len(parameters) > 10:
self.coupled_model(forcingtotal, doeclim_out, temperatures,
model, parameters)
model = gmsl_model.gmsl_model(theta, temperatures,
deltat) if self.alpha.value else []
lp_new = self.logp(theta_new, deltat, temperatures, model)
lq = lp_new - lp
lr = np.log(np.random.uniform(0, 1))
#print(lr, lq)
if (lr < lq):
theta = theta_new
lp = lp_new
accepts += 1
if lp > lp_max:
theta_best = theta
lp_max = lp
mcmc_chains[i] = theta
return mcmc_chains, accepts / N * 100
def chain(self, deltat, N=10000):
theta = np.array(self.values)
print('Initial estimate for parameters -', theta)
forcingtotal = forcing_total.forcing_total(
forcing=self.forcing,
alpha_doeclim=self.values[7],
l_project=False,
begyear=self.mod_time[0],
endyear=np.max(self.mod_time))
doeclim_out = doeclimF.doeclimF(forcingtotal,
self.mod_time,
S=self.values[5],
kappa=self.values[6])
temp_out = np.array(
(doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008), "temp"]).tolist())
ocheat = np.array((doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008),
'ocheat.mixed']).tolist())
ocheat = ocheat[self.offset[0]:self.offset[0] + 44]
temp_out += self.T_0
gmsl_out = gmsl_model.gmsl_model(theta, temp_out, 1)
lp = self.logp(theta, deltat, self.dfTemperature, gmsl_out, ocheat)
theta_best = theta
lp_max = lp
theta_new = [0.] * len(theta)
accepts = 0
mcmc_chains = np.zeros((N, (len(theta))))
step = np.diag(self.stepsizes)
sd = 2.38**2 / len(theta)
#step = 0.5 * step
#Check if converged. If not keep running.
print(N)
for i in (range(N)):
if i > 1 and not i % 1000:
pd.DataFrame(mcmc_chains).to_csv('array_uncorr.csv')
#print(i)
if i > 500: step = self.update_cov(mcmc_chains[:i], sd, len(theta))
if not i % 200: print(accepts)
theta_new = list(np.random.multivariate_normal(theta, step))
theta_new[5:] = [3.1, 3.5, 1.1, -0.06, 0.1, 0.55, 3, 0.5]
#temp_out, heatflux_mixed_out, heatflux_interior_out, gmsl_out = \
#doeclim_gmsl(asc = theta[7], t2co_in = theta[5], kappa_in=theta[6], alphasl_in = theta[0], Teq = theta[1], SL0 = theta[2], forcing=self.forcing)
#temp_out += theta[8]
#ocheat = self.flux_to_heat(heatflux_mixed_out, heatflux_interior_out)
#ocheat = ocheat[self.offset[0]:self.offset[0]+ 44]
doeclim_out = doeclimF.doeclimF(forcingtotal,
self.mod_time,
S=theta[5],
kappa=theta[6])
temp_out = np.array(
(doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008),
"temp"]).tolist())
temp_out += theta[8]
ocheat = np.array((doeclim_out.loc[(doeclim_out["time"] >= 1880) &
(doeclim_out["time"] <= 2008),
'ocheat.mixed']).tolist())
gmsl_out = gmsl_model.gmsl_model(theta, self.dfTemperature, 1)
ocheat = ocheat[self.offset[0]:self.offset[0] + 44]
lp_new = self.logp(theta_new, deltat, temp_out, gmsl_out, ocheat)
if np.isinf(lp_new):
mcmc_chains[i, :] = theta
continue
#print(lp, lp_new, 'difference')
lq = lp_new - lp
lr = np.log(np.random.uniform(0, 1))
#print(lr, lq)
if (lr < lq):
theta = theta_new
lp = lp_new
accepts += 1
#print('NEW VALUE')
if lp > lp_max:
theta_best = theta
lp_max = lp
mcmc_chains[i, :] = theta
return mcmc_chains, accepts / N * 100