def test_target_timeseries_pred():
np.random.seed(123)
goal = np.array([-10.549816, -0.657204, -4.163044, -12.341982, -6.353174])
proxy_ts = np.array([0.2831, 0.2856, 0.2832, 0.2854, 0.3081])
alpha_now = 0.3584
beta_now = 0.0054
tau2_now = 0.0016
prior_pars = {'mu': np.array([0.0535] * 5), 'inv_cov': np.eye(5) * 0.0278}
victim = target_timeseries_pred(alpha_now, beta_now, tau2_now, proxy_ts,
prior_pars)
np.testing.assert_allclose(victim, goal, atol=1e-4)
def predict_seatemp_analog(tex, prior_std, temptype, search_tol, prior_mean=None, nens=5000, progressbar=True):
"""Predict sea temperature with TEX86, using the analog method
Parameters
----------
tex : ndarray
n-length array of TEX86 observations from a single location.
prior_std : float
Prior standard deviation for sea temperature (°C).
temptype : str
Type of sea temperature desired. Either 'sst' for sea-surface or 'subt'.
search_tol: float
Tolerance for finding analog locations. Comparison is between the mean
of dats and the mean tex value within each large gridcell.
prior_mean : float
Prior mean for sea temperature (°C).
nens : int
Size of MCMC ensemble draws to use for calculation.
progressbar: bool
Whether or not to display a progress bar on the command line. The bar
shows how many analogs have been completed.
Returns
-------
output : Prediction
Raises
------
EnsembleSizeError
"""
draws = get_draws(temptype)
tex_obs = get_tex(temptype)
nd = len(tex)
ntk = draws.alpha_samples_comp.shape[1]
if ntk < nens:
raise EnsembleSizeError(ntk, nens)
latlon_match, val_match = tex_obs.find_within_tolerance(x=tex.mean(),
tolerance=search_tol)
n_locs_g = len(latlon_match)
prior_par = {'mu': np.ones(nd) * prior_mean,
'inv_cov': np.eye(nd) * prior_std ** -2}
n_latlon_matches = len(latlon_match)
indices = range(n_latlon_matches)
if progressbar:
indices = tqdm(indices, total=n_latlon_matches)
preds = np.empty((nd, n_locs_g, nens))
for kk in indices:
latlon = latlon_match[kk]
alpha_samples, beta_samples = draws.find_alphabeta_near(*latlon)
for jj in range(nens):
a_now = alpha_samples[jj]
b_now = beta_samples[jj]
t2_now = draws.tau2_samples[jj]
preds[:, kk, jj] = target_timeseries_pred(alpha_now=a_now,
beta_now=b_now,
tau2_now=t2_now,
proxy_ts=tex,
prior_pars=prior_par)
output = Prediction(ensemble=preds,
temptype=temptype,
prior_mean=prior_mean,
prior_std=prior_std,
analog_gridpoints=latlon_match)
return output
def predict_seatemp(tex, lat, lon, prior_std, temptype, prior_mean=None, nens=5000):
"""Predict sea temperature with TEX86
Parameters
----------
tex : ndarray
n-length array of TEX86 observations from a single location.
lat : float
Site latitude from -90 to 90.
lon : float
Site longitude from -180 to 180.
prior_std : float
Prior standard deviation for sea temperature (°C).
temptype : str
Type of sea temperature desired. Either 'sst' for sea-surface or 'subt'.
prior_mean : float or None, optional
Prior mean for sea temperature (°C). If 'None', the prior mean is found
by searching for a "close" value in observed sea temperature records.
nens : int
Size of MCMC ensemble draws to use for calculation.
Returns
-------
output : Prediction
Raises
------
EnsembleSizeError
"""
draws = get_draws(temptype)
obs = get_seatemp(temptype)
nd = len(tex)
ntk = draws.alpha_samples_comp.shape[1]
if ntk < nens:
raise EnsembleSizeError(ntk, nens)
if prior_mean is None:
close_obs, close_dist = obs.get_close_obs(lat=lat, lon=lon)
prior_mean = close_obs.mean()
alpha_samples_comp, beta_samples_comp = draws.find_alphabeta_near(lat=lat, lon=lon)
tau2_samples = draws.tau2_samples
grid_latlon = draws.find_nearest_latlon(lat=lat, lon=lon)
prior_par = {'mu': np.ones(nd) * prior_mean,
'inv_cov': np.eye(nd) * prior_std ** -2}
preds = np.empty((nd, nens))
for jj in range(nens):
preds[:, jj] = target_timeseries_pred(alpha_now=alpha_samples_comp[jj],
beta_now=beta_samples_comp[jj],
tau2_now=tau2_samples[jj],
proxy_ts=tex,
prior_pars=prior_par)
# TODO(brews): Consider a progress bar for this loop.
output = Prediction(ensemble=preds,
temptype=temptype,
latlon=(lat, lon),
prior_mean=prior_mean,
prior_std=prior_std,
modelparam_gridpoints=[tuple(grid_latlon)])
return output