Sparse matrix data#

Sparse matrix format can greatly reduce size on disk in case most of the matrix is occupied with zeros. For every input detector a single file with .npz extension is written. It contains:

  • shape of detector data: tuple of 3 numbers holding nx, ny and nz

  • tuple of 3 numpy arrays with X, Y and Z coordinates, pointing to non-zero data cells

  • plain, 1D numpy array with non-zero data values

Such format is similar to a so-called COO-rdinate format for 2D sparse matrices. Here it is used for any kind of data, with dimensionality between 0 and 3.

Output file is saved in uncompressed NPZ numpy data format, for details see https://docs.scipy.org/doc/numpy/reference/generated/numpy.savez.html

An example usage#

Conversion is done using standard command:

convertmc sparse --many "*.bdo"

Assuming that we had “dose” detector output saved to dose0001.bdo, dose0002.bdo and dose0003.bdo, we should expect to get dose.npz as an output file.

Data reconstruction#

Sparse data can be extracted from NPZ file with following Python code:

import numpy as np

# load contents of NPZ file:
npzfile = np.load(filename)
data = npzfile['data']   # plain 1-D numpy array
indices = npzfile['indices']  # 3-elements tuple with array of coordinates
shape = npzfile['shape'] # 3-element shape tuple

# reconstruct normal form of matrix from sparse data
result = np.zeros(shape)
for ind, dat in zip(indices, data):
    result[tuple(ind)] = dat

Threshold value#

Additionally a threshold can be specified to discard noisy data. In this case, all data cells with absolute value less or equal than threshold will be treated as zeros. Please note that after reconstructing from sparse format we won’t get the original data, as values lower than threshold will be overwritten with zeros. An example usage:

convertmc sparse --many "*.bdo" --threshold 1e-7