This is the Python API of KDSource package for Monte Carlo radiation calculations.

KDSource models distribucional particle sources thru Kernel Density Estimation (KDE), based on particle lists with MCPL format. It allows to couple different simulations implementing variance reduction, helping Monte Carlo calculations in big systems.

This API allows creating, optimizing, analyzing, plotting, and saving KDSource sources. Internally, it uses the KDEpy library for (adaptive) KDE. The optimization consists in automatic selection of KDE bandwidth. For this purpose, the following methods are included:

• Silverman's Rule: Simple, fast and not very precise. It is based only on the number of particles and geometry dimension.
• K Nearest Neighbors: For each sample, computes its bandwidth as the distance to the k-th neighbor.
• Maximum Likelihood Cross-Validation: Builds a grid of bandwidths, and computes the cross-validation likelihood score for each, which is an indicator of the quality of the estimation. Then chooses the bandwidth that maximizes it. It requires a seed bandwidth, which can be provided by any of the previous methods.

# Create and fit KDSource

plist = kds.PList("mcplfile.mcpl")      # PList: Wrapper for MCPL file
geom = kds.Geometry([kds.Lethargy(),     # Geometry: define metrics for variables
					kds.SurfXY(),
					kds.Polar()])
s = kds.KDSource(plist, geom, bw='silv') # Create KDSource
s.fit(N=1E4)                           # Fit with N particles, optimizing bandwidth

# Create plots

# Energy plot
EE = np.logspace(-3,0,30)
fig,[scores,errs] = s.plot_E(EE)

# Custom variable 1D plot
tt = np.linspace(0,180,30)
fig,[scores,errs] = s.plot_integr('theta', tt)

# Custom variable 2D plot
xx = np.linspace(-10,10,30)
yy = np.linspace(-10,10,30)
fig,[scores,errs] = s.plot2D_integr(['x','y'], [xx,yy])

# Save KDSource to XML file
s.save("xmlfile.xml")