lidisa is a lightweight implementation of direct sampling in Python. This is a simplified version of the algorithm described in The Direct Sampling method to perform multiple‐point geostatistical simulations published by Mariethoz, Renard and Straubhaar in Water Resources Research (2010). It requires only numpy and numba as dependencies and makes use of the numba just-in-time compiler to significantly speed up sampling. lidisa is able to conduct both conditional and unconditional sampling of new images based on training data. Using it is simple:
training_image = ...
simulator = lidisa.dsampler(training_image, iterations=5)
simulations = [x for x in simulator]
Currently, lidisa supports only categorical-valued images. Images with continuous values will be supported in a future release. See the attached Jupyter notebook for a detailed example workflow using lidisa to conduct stochastic simulation. To install this repository, clone it and run python setup.py install from within the directory.
For questions, comments or concerns please email Christopher Krapu at ckrapu@gmail.com.