Functions to construct distance curves used for class prior estimation
makeCurvesFromDistanceMatrix(dist_matrix, curves, mixtureInstanceRemaining)
Construct multiple distance curves, using the precomputed distances
Arguments:
- dist_matrix : float[num_component_instances, num_mixture_instances]
dist_matrix[i,j] contains the distance between component instance i and mixture instance j
- curves : float[num_curves_to_average, num_mixture_instances]
matrix to fill with distance curves (passed initialized matrix because jit
in no python mode can't create matrices)
- mixtureInstanceRemaining : boolean[num_curves_to_average, num_mixture_instances]
boolean matrix indicating whether the given mixture instance should be
considered when constructing the curve at that iteration.
At each iteration, the mixture point that is closest to the sampled component
instance is removed from consideration in subsequent iterations.
Pass matrix initialized to all True in order for all mixture instances to be
considered in curve construction.
makeCurve()
Construct the distance curve used to estimate the class prior
of the distribution from which the mixture instances were sampled
Arguments:
- compInstances : float[num_component_instances, dim]
instances sampled from the component distribution
- mixInstances : float[num_mixture_instances, dim] in range[0,1]
instances sampled from the mixture distribution
- metric : string or callable
only used if gpu==False
see scipy.spatial.distance.cdist for details
- num_curves_to_average : int : default 25
repeat the curve construction process this number of times, averaging over all curves
- quantiles : float[n_quantiles] : default np.arange(0,1,.01)
after averaging over all curves, use these quantiles of the averaged curve
as the final distance curve
plotCurve(curve)
Plot the given distance curve