Release 0.6.3

[statrita2004]

Combining #94 and #95:

• Fixed the MCMC routine (used with BSL and others) when a starting point for the chain is passed. If a 1d array or list was given, in fact, parameter shape at the end of inference was not correct and the diagnostic functions did not work properly.
• Added a DrawFromPrior inference scheme, with similar API as the other inference schemes and is used to sample from prior with no need to specify observation. That also has another method which can be used to generate simulation-parameter pairs to be used for learning the automatic summary statistics utilities. I've added an example showing how to do that, and discussed that in the docs page.
• Added a method to the Journal which allows to resample the posterior samples (bootstrap and subsample). That generates a new journal which is returned by the method.
• Added a new GenerateFromJournal class which allows to generate simulations from a given model using parameter values stored in a journal. Together with the previous method this allows to perform predictive check.
• Some reformatting to the Statistics; specifically, I've added the capability to standardize the different statistics by dividing them by their standard deviation on a set of reference simulations. Those simulations need to be provided at initialization of the statistics, at which point the standard deviation is computed. Then, each time the .statistics method is called, the resulting statistics are normalized by the standard deviation. Can be useful to use in cases in which the summaries have wildly varying magnitude.
• Added the Statistics learning with exponential family based on Score Matching.
• Refactored some code in the Journal class and its uses in the InferenceMethods: the add_opt_values method was outdated and used for two different thing in two difference InferenceMethod's. I have now added methods to store simulations and covariance matrices in the Journal.
• Added option to not store the simulated dataset in the Journal for SMCABC. That breaks the possibility of restarting inference from the journal but greatly reduces journal size.
• Added link to Youtube video presenting ABCpy in README
• Added an example showing how to save to disk and re-load the learned neural network statistics.
• Some minor fixes in docs and tests.

[codecov-commenter]

Codecov Report

Merging #95 (3ee238d) into master (3ee0aee) will increase coverage by 1.18%.
The diff coverage is 89.25%.

@@            Coverage Diff             @@
##           master      #95      +/-   ##
==========================================
+ Coverage   83.72%   84.90%   +1.18%     
==========================================
  Files          27       27              
  Lines        5424     6143     +719     
  Branches      993     1143     +150     
==========================================
+ Hits         4541     5216     +675     
- Misses        600      623      +23     
- Partials      283      304      +21     

Impacted Files	Coverage Δ
abcpy/continuousmodels.py	82.37% <ø> (ø)
abcpy/distances.py	89.50% <ø> (ø)
abcpy/statisticslearning.py	80.55% <78.18%> (+2.16%)	⬆️
abcpy/NN_utilities/networks.py	85.71% <87.28%> (+6.05%)	⬆️
abcpy/NN_utilities/trainer.py	80.24% <91.66%> (+1.86%)	⬆️
abcpy/inferences.py	88.13% <93.13%> (+0.45%)	⬆️
abcpy/output.py	87.82% <96.26%> (+4.12%)	⬆️
abcpy/NN_utilities/utilities.py	95.50% <96.82%> (+3.19%)	⬆️
abcpy/NN_utilities/algorithms.py	62.74% <100.00%> (+9.80%)	⬆️
abcpy/NN_utilities/losses.py	100.00% <100.00%> (ø)
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3ee0aee...3ee238d. Read the comment docs.