Pypesto

------------------

Major:

* Modularize import, to import optimization, sampling and profiling
separately (413)

Minor:

* Bug fixes in
* sampling (412)
* visualization (405)
* PEtab import (403)
* Hessian computation (390)

* Improve hdf5 error output (409)
* Outlaw large new files in GitHub commits (388)

..........

------------------

Objective

* Write solver settings to stream to enable serialization for distributed
systems (308)

* Refactor objective function (347)
* Removes necessity for all of the nasty binding/undbinding in AmiciObjective
* Substantially reduces the complexity of the AggregatedObjective class
* Aggregation of functions with inconsistent sensi_order/mode support
* Introduce ObjectiveBase as an abstract Objective class
* Introduce FunctionObjective for objectives from functions

* Implement priors with gradients, integrate with PEtab (357)
* Fix minus sign in AmiciObjective.get_error_output (361)
* Implement a prior class, derivatives for standard models, interface with
PEtab (357)
* Use `amici.import_model_module` to resolve module loading failure (384)

Problem

* Tidy up problem vectors using properties (393)

Optimization

* Interface IpOpt optimizer (373)

Profiles

* Tidy up profiles (356)
* Refactor profiles; add locally approximated profiles (369)
* Fix profiling and visualization with fixed parameters (393)

Sampling

* Geweke test for sampling convergence (339)
* Implement basic Pymc3 sampler (351)
* Make theano for pymc3 an optional dependency (allows using pypesto without
pymc3) (356)
* Progress bar for MCMC sampling (366)
* Fix Geweke test crash for small sample sizes (376)
* In parallel tempering, allow to only temperate the likelihood, not the prior
(396)

History and storage

* Allow storing results in a pre-filled hdf5 file (290)
* Various fixes of the history (reduced vs. full parameters, read-in from file,
chi2 values) (315)
* Fix proper dimensions in result for failed start (317)
* Create required directories before creating hdf5 file (326)
* Improve storage and docs documentation (328)
* Fix storing x_free_indices in hdf5 result (334)
* Fix problem hdf5 return format (336)
* Implement partial trace extraction, simplify History API (337)
* Save really all attributes of a Problem to hdf5 (342)

Visualization

* Customizable xLabels and tight layout for profile plots (331)
* Fix non-positive bottom ylim on a log-scale axis in waterfall plots (348)
* Fix "palette list has the wrong number of colors" in sampling plots (372)
* Allow to plot multiple profiles from one result (399)

Logging

* Allow easier specification of only logging for submodules (398)

Tests

* Speed up travis build (329)
* Update travis test system to latest ubuntu and python 3.8 (330)
* Additional code quality checks, minor simplifications (395)

-------------------

* Tidy up and speed up tests (265 and others).
* Basic self-implemented Adaptive Metropolis and Adaptive Parallel Tempering
sampling routines (268).
* Fix namespace sample -> sampling (275).
* Fix covariance matrix regularization (275).
* Fix circular dependency `PetabImporter` - `PetabAmiciObjective` via
`AmiciObjectBuilder`, `PetabAmiciObjective` becomes obsolete (274).
* Define `AmiciCalculator` to separate the AMICI call logic (required for
hierarchical optimization) (277).
* Define initialize function for resetting steady states in `AmiciObjective`
(281).
* Fix scipy least squares options (283).
* Allow failed starts by default (280).
* Always copy parameter vector in objective to avoid side effects (291).
* Add Dockerfile (288).
* Fix header names in CSV history (299).

Documentation:

* Use imported members in autodoc (270).
* Enable python syntax highlighting in notebooks (271).

-------------------

* Add typehints to global functions and classes.
* Add `PetabImporter.rdatas_to_simulation_df` function (all 235).
* Adapt y scale in waterfall plot if convergence was too good (236).
* Clarify that `Objective` is of type negative log-posterior, for
minimization (243).
* Tidy up `AmiciObjective.parameter_mapping` as implemented in AMICI now
(247).
* Add `MultiThreadEngine` implementing multi-threading aside the
`MultiProcessEngine` implementing multi-processing (254).
* Fix copying and pickling of `AmiciObjective` (252, 257).
* Remove circular dependence history-objective (254).
* Fix problem of visualizing results with failed starts (249).
* Rework history: make thread-safe, use factory methods, make context-specific
(256).
* Improve PEtab usage example (258).
* Define history base contract, enabling different backends (260).
* Store optimization results to HDF5 (261).
* Simplify tests (263).

Breaking changes:

* `HistoryOptions` passed to `pypesto.minimize` instead of `Objective` (256).
* `GlobalOptimizer` renamed to `PyswarmOptimizer` (235).

-------------------

* Rewrite AmiciObjective and PetabAmiciObjective simulation routine to directly use
amici.petab_objective routines (209, 219, 225).
* Implement petab test suite checks (228).
* Various error fixes, in particular regarding PEtab and visualization.
* Improve trace structure.
* Fix conversion between fval and chi2, fix FIM (all 223).