AFL.automation.instrument.GPInterpolator

Classes

ClusteredGPs(dataset)	This is the wrapper class that enables scattering interpolation within the classifier labels.
Interpolator(dataset)

class AFL.automation.instrument.GPInterpolator.Interpolator(dataset)

__init__(dataset)

get_defaults()

set_defaults(default_dict)

load_data(dataset=None)

standardize_data()

construct_model(kernel=None, noiseless=False, heteroscedastic=False)

train_model(kernel=None, optimizer=None, noiseless=False, heteroscedastic=False, tol=0.0001, niter=21)

predict(X_new=[[0, 0, 0], [1, 2, 3]])

Returns the simulated scattering pattern given the specified coordinates and polynomial type if reduced

print_diagnostics()

class AFL.automation.instrument.GPInterpolator.ClusteredGPs(dataset)

This is the wrapper class that enables scattering interpolation within the classifier labels. It presupposes that the dataset or manifest contains a datavariable called labels and that it is

__init__(dataset)

get_defaults()

returns a list of dictionaries corresponding to the default data pointers for each GP model

set_defaults(default_dict)

check the default params, but these are pointers to the values in the dataset

load_datasets(gplist=None)

This function instantiates the X_raw, Y_raw data with the appropriate filtering given the defaults dictionaries

define_domains(gplist=None, alpha=0.1, buffer=0.01)

define domains will generate the shapely geometry objects for the given datasets X_raw data points.

unionize(gplist=None, geomlist=None, dslist=None, buffer=0.01)

determines the union of and indices of potentially conflicting domains. Creates the new indpendent_GPs list that corresponds.

Note! buffer is important here! if a cluster contains a single point or a line, there is a bunch of stuff that breaks. specifying a non-zero buffer will force these lesser dimensional objects to be polygonal and help with all the calculations. a large buffer will potentially merge polygons though. exercise caution

train_all(kernel=None, optimizer=None, noiseless=True, heteroscedastic=False, niter=21, tol=0.0001, gplist=None)

predict(X_new=None, gplist=None, domainlist=None, shapely_domains=False)

Returns posterior mean and uncertainty of the pattern for the model closest to the input coordinate. Note that the coordinate should be in natural units specified by the Interpolator.predict function

print_diagnostics()