Changes
- Previously, we had been using overlapping samples of displacements (i.e. the blue and orange observations in the figure below). However, this puts us on thin statistical ice and we were using a “fudge factor” to account for this (which in turn leads to a slightly overestimated variance in diffusion coefficient. We have changed this to only use non-overlapping samples (i.e. the blue and green observations in the figure below). This is more statistically sound and gives a more accurate estimate of the variance in the diffusion coefficient. (https://github.com/bjmorgan/kinisi/commit/03a6edc86e9975f1e9ace93396df5ed5ac06396b)
- Removal of the parser keyword arguments `min_obs` and `ndelta_t` these were artefacts of the old overlapping sampling approach and therefore have been removed. (https://github.com/bjmorgan/kinisi/commit/03a6edc86e9975f1e9ace93396df5ed5ac06396b)
- Now by default, the number of Δt points will reflect that in the simulation (previously there was a reduced sampling approach that has been removed). (https://github.com/bjmorgan/kinisi/commit/03a6edc86e9975f1e9ace93396df5ed5ac06396b)
- Thinning and the ability to add a `random_state` to the Markov chain Monte Carlo process have been introduced (see the keyword arguments for the diffusion params, `thin` and `random_state`). (https://github.com/bjmorgan/kinisi/commit/03a6edc86e9975f1e9ace93396df5ed5ac06396b)