Pyabc

..........

-----

* Acceptance transferred to an Acceptor object to allow for more
flexibility (i.e. not only on a single comparison as per default).
* This acceptor is passed to the ABCSMC object.
* Update of distance and epsilon synchronized after each iteration and moved
to update() methods.
* initialize() for DistanceFunction and Epsilon also called in load() method,
given a time point to initialize for, and made optional via a
require_initialize flag. This makes sure these objects are always correctly
initialized.
* PNormDistance and AdaptivePNormDistance (prev. WeightedPNormDistance)
improved to allow for more customization.
* ABCSMC.set_data() method removed.
* API breaks for DistanceFunction, Epsilon, Model.

------

* Implementation of adaptive distances feature. Distance functions can adapt
via an update() method.
* In particular add WeightedPNormDistance (special case:
WeightedEuclideanDistance). Also add non-weighted versions.
* Simplify Sampler.sample_until_n_accepted interface.
* Extend Sampler class to allow for customization, e.g. by the distance
functions.
* Generalize MedianEpsilon to QuantileEpsilon.
* Make Viserver work with latest bokeh version.

------

* Add batch sampling now also to the REDIS evaluation parallel sampler
(dynamic scheduling)

------

* Bug fix. Fix a race condition in the redis evaluation parallel sampler
(dynamic scheduling). An error occured if a worker tried to start to work
on a population after the other workers had already terminated the
population.

------

* Minor bug fix. Ensure that the multicore samplers raise an Exception if
an Exception occurs in the worker processes.
* Clarify that weighted distances are not normalized in case of having more
than a single simulation per proposed parameter.
Also add corresponding tests.
* Add n_worker method to the RedisEvalParallelSampler to enable querying of
the number of connected workers.
* Add in-memory database support. Useful, e.g., for benchmarking on slow
filesystems or with rather slow network connections.