* Minor documentation tweaks and corrections.
* Grammar in evo.ge.Ge is always transformed into the evolutils.gramar.Grammar as it is needed by the subtree mutation.
* Mutations are set up similarly like crossover using the setup_mutation() method.
* Changed annotations from subtree crossover are stored in the offsprings.
* Introduced subrree mutation, variable mutation.
* evo.utils.grammar.Grammar.to_tree() now can take extra argument allowing to override the start symbol.
* Annotations returned from evo.utils.grammar.Grammar.to_tree() are now lists instead of tuples.

Extraction of fitness from error in evo.sr.MultiGeneGeSrFitness moved into separate method that can be easilly overriden.

* BSF solution and its data is pretty-printed after a run of evo.ge.Ge finishes.
* End of evo.ge.support.RampedHalfInitializer intialization run is logged.
* Adjusted .gitignore to ignore logging.yaml wich serves only for manual testing purposes.

* New method get_minimum_expansion_depth() in evo.utils.grammar.Grammar that returns the minimum depth of a derivation tree that forms a fully expanded grammar.
 * Unit test created too.

* Deleted evo.tests.support and test_grammar moved to evo.tests.utils.
* test_grammar fixed to reflect recent changes in grammar api (annotations are tuples, incomplete expansion results to non-None, though incomplete, annotations).

Individual is passed to the parsing methods too so that subclasses can store some information to it.

* Different error handling in evo.sr.MultiGeneGeSrFitness
** A fitness value when an error occurs (error fitness) is specified separately from the unfinished-transcription fitness
** Errors handled by the error fitness can be specified manually (ZeroDivisionError and FloatingPointError are handled by default).
** numpy.linalg.linalg.LinAlgError is handled by the error fitness too (when SVD does not converge)
* New class evo.sr.MultiGeneGe
** Derives from evo.ge.Ge.
** Implements special multi-gene operators.
* Fixed bug in evo.ge.support.RampedHalfHalfInitializer.
* Adjustments of evo.ge.Ge
** Setting up the crossover algorithm removed from the constructor to a separate method which can be overriden.
** All crossovers unset the annotations.

* Methods to_tree and to_text of evo.utils.grammar.Grammar return the annotations as tuples (in order to be hashable).
* evo.ge.support.RandomWalkInitializer and RampedHalfHalfInitializer can add extra "useless" codons after the main codon sequence.

* The features data is no longer put inside this class - the responsibility is on the apply_gene_phenotype() method.

* The stored sequence in to_tree() is adapted accordingly so that the trees/texts generated with it will match this initial tree.

* The best-so-far is properly tested in extract_elites() too.

* The data field in evo.Individual (and its subclasses) made private (by renaming to _data).
** It is not to be manipulated directly but through new methods get_data() and set_data().
* Variable crossover in evo.ge.Ge. Works with any supported ground crossover methods.
* Fastu mutation by sampling the index of the codon to be mutadet from a geometric distribution, lowering the number of random number generator calls.

* If the genotype of either parent is too short that it did not expand any rule the subtree crossover is not performed (there are no offsprings of such parents).
* Annotations are properly cloned in evo.ge.support.CodonGenotypeIndividual.clone().
* All acquired annotations are returned, even if the expansion was not fully finished.

* Missed return statement.

* Created evo.ge.support.RampedHalfHalfInitializer.

* Grammar rules can tell which choices are non-terminal (for full & grow methods).
* Grammar can generate derivation trees with lower depth limit too.

* individual was added as an argument to method evo.GeFitness.make_phenotype so that it can record some info to the original individual.

* .gitignore enhanced with content generated by gitignore.io using keywords: Python, PyCharm.
* Very basic README.rst content: simple installation instructions.

* Created class evo.sr.MultiGeneGeSrFitness
** Converts the low-level grammar to the multi-gene grammar - no need to manually design the multi-gene grammar.
** Performs the evaluation of multi-gene individuals.
** Takes care of comparing fitness values (not sorting).
* New methods for trees (and unit tests).
* Added numpy dependency.
* Added numpy intersphinx link.

* Created a SR module (sr.py) skeleton which will contain symbolic-regression-specific implementations of GP.

* Common base class GeFitness for fitnesses to work with GE.
* Two abstract subclasses of GeFitness
** GeTreeFitness that decodes the genome to a derivation tree (which is then further processed)
** GeTextFitness that decodes the genome to a text string (which is then further processed).

* Completed unit test.
* Finished implementation.

Fixes #1

* Implemented the crossover method.
* Created a unit test skeleton.

#1

* Individual classes and Initializers moved from evo.ge to evo.ge.support.
* Fixed ReST warnings from docstrings.

* Encoding specification at the beginning of each .py file.
* Docstring as the first command of each .py file (except setup.py)
* __author__ declaration in each .py file (except setup.py).
* String literals are always enclosed in apostrophes (').
* Docstrings are always enclosed in triple double-quotes (""").
* evo.support renamed to evo.utils.

#1