Anticpy

Unnecessary comments have been removed to tidy up the code. This release represents the version that is discussed throughout the corresponding PhD thesis. Main functionalities and algorithms discussed therein are reliably working.

**These functionalities include:**

- Bayesian Langevin estimation via MCMC, MAP, and binned MAP estimation,
- Non-Markovian Bayesian Langevin estimation via MCMC and MAP,
- Trend change point analysis with marginalization option for the joint probability of change point configurations,
- Trend change point analysis with optimized memory management and on-the-fly construction of individual change point configurations,
- Simplified dominant eigenvalue estimation as procedural functions, including method parameter optimization and graphical output,
- video animation for the Bayesian Langevin estimation.

Apart from these main functionalities there are several further options that are not tested in detail.

**Functionalities implemented that may be tested in future include:**

- The package provides several options for approximations of drift and diffusion functions. The current version ist tested for linear or third-order polynomial drift functions with constant noise. In the non-Markovian case, the second process is an Ornstein-Uhlenbeck process that replaces the Wiener process in the Markovian version. Other combinations should work but are not tested in detail.
- The anaimation option for the Bayesian Langevin estimation reliably works. However, a preliminary version for animation of non-Markovian results is provided, but not tested yet.

With this release various functionalities in the `trend_extrapolation` package are optimized regarding computation times and memory management:

1. The `numpy.roll` function is replaced by fancy indexing during accessing the change point configurations in the batched change point analysis. This notably reduces the computation time.
2. The `numpy.roll` function is avoided during the marginalization of the probability density functions of the change point positions.
3. The parallel workflow during the marginalization of the probability density functions of the change point positions is optimized. The number of jobs per parallel worker is optimally distributed before starting the parallel processing pool. Each worker gets the same number of jobs if possible. If this is not possible, one worker additionally processes the remainder.
4. For marginalizing the joint probability density function of the change point configurations, a `queue_management` is implemented that restarts the pool by a definable number of times with a subset of jobs to avoid memory errors.

This release acts as a preparation for the final PhD release that is planned to represent the documented version in the corresponding PhD thesis.

Add a further print output correction.

Two minor adaptions to the print output are included:

1. The completion_control value which is used to control the correct termination of cp_scan(...), fit(...), and compute_CP_pdfs(...) is reset to zero in compute_CP_pdfs(...) to avoid wrong outputs if one of the former methods is used before the latter.
2. The print output of the batched CP fit is corrected for the singular or plural of "round".

This post-release is related to the computation of the marginal CP PDFs. For a high number of CP configurations, the multiprocessing pool allocates excessive amounts of memory to prepare all jobs. Therefore, the option

- `queue_managment` is added which allows for opening and closing of multiprocessing pools per batch. This resolves the memory issue and guarantees feasible computation times.

**Adaptations:**

1. Remove window size output for MAP estimation procedures.
2. Adapt the variable name batchsize in batched_configs(...) to batch_size for congruence.

**Bug Fixes:**

1. Correct multiprocessing options of the emcee MCMC sampling in perform_resilience_scan(...).
2. Correct normalization of the marginal CP PDFs in compute_CP_pdfs(...). The marginalized PDFs now individually sum up to one.

v.0.0.8.post1
**Minor Enhancements:**

- The _window_shift_ option of the _RocketFastResilienceScan_ wrapper is enhanced. Instead of the integer value for an equidistant window shift which is already known from _LangevinEstimation_ and _NonMarkovEstimation_, the parameter can be defined as one dimensional numpy array of integers. The entries correspond to specific window shifts that are executed by the workers before computing the corresponding windows. This can help to easily fill up missing values of a first parallel calculation or to pick out specific windows of interest in general.

**Bug Corrections:**

- After the last release, the _nburn_ parameter passing was corrected for _LangevinEstimation_ and _NonMarkovEstimation_. Some bug remained for the _LangevinEstimation_ class. It is corrected in this post release.