def fitGLSRobust(geodataframe,
variogram_object,
num_iterations=20,
distance_threshold=1000000):
"""
Fits a GLS model iterating through the residuals of the previous GLS.
After estimating the parameters the method uses the new residuals as input.
Recalculates the empirical variogram, refits the theoretical variogram, recalculates de Covariance Matrix.
This is done `num_iterations` times.
note: the geodataframe needs to have a column called `residuals`
"""
lrsq = []
lparams = []
lpvals = []
lconf_int = []
for i in range(num_iterations):
logger.info("Building Spatial Covariance Matrix")
CovMat = buildSpatialStructure(geodataframe, variogram_object.model)
logger.info("Calculating GLS estimators")
results, resum = calculateGLS(geodataframe, CovMat)
n_obs = results.nobs
rsq = results.rsquared
params = results.params
pvals = results.pvalues
conf_int = results.conf_int()
logger.info("RESULTS::: n_obs: %s, r-squared: %s, {%s,%s,%s}" %
(n_obs, rsq, params.to_json(), pvals.to_json(),
conf_int.to_json()))
lrsq.append(rsq)
lparams.append(params)
lpvals.append(pvals)
lconf_int.append(conf_int)
geodataframe.residuals = results.resid
envelope = variogram_object.envelope
variogram_object = tools.Variogram(
geodataframe,
'residuals',
using_distance_threshold=distance_threshold,
model=variogram_object.model)
variogram_object.envelope = envelope
#variogram_object.empirical = empirical
logger.info("Recalculating variogram")
variogram_object.calculateEmpirical()
logger.info("Refiting Theoretical Variogram")
tt = variogram_object.fitVariogramModel(variogram_object.model)
logger.info("Variogram parameters: range %s, sill %s, nugget %s" %
(tt.range_a, tt.sill, tt.nugget))
resultspd = pd.DataFrame({
'rsq': lrsq,
'params': lparams,
'pvals': lpvals,
'conf_int': lconf_int
})
return (resum, variogram_object, resultspd, results)
def buildSpatialStructure(geodataframe, theoretical_model):
"""
wrapper function for calculating spatial covariance matrix
"""
secvg = tools.Variogram(geodataframe,
'logBiomass',
model=theoretical_model)
logger.info("Calculating Distance Matrix")
CovMat = secvg.calculateCovarianceMatrix()
return CovMat
def loadVariogramFromData(plotdata_path, geodataframe):
"""
Reads and instantiates a Variogram object using data stored in plotdata_path.
"""
## Read the empirical variogram
logger.info("Reading the empirical Variogram file")
thrs_dist = 1000000
## Change here with appropriate path for file
empirical_semivariance_log_log = plotdata_path
#### here put the hec calculated
logger.info(
"Instantiating a Variogram object with the values calculated before")
emp_var_log_log = pd.read_csv(empirical_semivariance_log_log)
gvg = tools.Variogram(geodataframe,
'logBiomass',
using_distance_threshold=thrs_dist)
gvg.envelope = emp_var_log_log
gvg.empirical = emp_var_log_log.variogram
gvg.lags = emp_var_log_log.lags
logger.info("Dropping possible Nans")
emp_var_log_log = emp_var_log_log.dropna()
vdata = gvg.envelope.dropna()
logger.info("Instantiating Model...")
#matern_model = tools.MaternVariogram(sill=0.34,range_a=100000,nugget=0.33,kappa=0.5)
whittle_model = tools.WhittleVariogram(sill=0.34,
range_a=100000,
nugget=0.0,
alpha=3)
logger.info("fitting %s Model with the empirical variogram" %
whittle_model.name)
gvg.model = whittle_model
tt = gvg.fitVariogramModel(whittle_model)
logger.info("Model fitted")
return (gvg, tt)