Visual-graph-datasets

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

- Added some more graph utility functions such as functions to extract sub graphs, add and remove nodes and
to identify connected regions of a graph.
- Added documentation for the ``ColorProcessing`` class
- Changed the ``ColorProcessing.visualize_as_figure`` method to now also accept external graph dict parameter
and external node_positions array.
- Modified the ``generate_molecule_dataset_from_csv.py`` experiment so that it is now possible to optionally
define a indices blacklist of elements that should be skipped during processing.
- Moved the dependencies to the most recent version of RDKit. This seems to have fixed the issue of the molecule
image generation occasionally crashing with a segmentation fault.
- Added the ``generic`` graph type. This is a graph type that can be used to represent any kind of graph
that cannot be associated with any kind of specific domain. Added the ``GenericProcessing`` class which
can be used to process these generic graphs.
- Modifed the ``colors_layout`` function such that it is possible to pass a partially defined list of
node_positions as an argument, such that the positions of some nodes can be fixed during the layouting.

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

- Added back in the dictionaries defining the alternative versions for the node and edge importance plotting

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

- Fixed a minor issue where the datasets folder was not created during the ``config`` initialization which
has led to errors when trying to download a dataset.

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

- Added a utility function to count how often a subgraph motif appears in a larger graph
- Added experiment ``analyze_color_graph_dataset.py`` to analyze the properties of color graph based datasets

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

- Added the experiment ``profile_molecule_processing.py`` to profile and plot the runtime of the different
process components that create a visual graph dataset element with the aim of identifying the source of the
runtime degradation bug.
- Fixed the runtime degradation / memory leak issue in ``generate_molecule_dataset_from_csv.py``. It seems like the
problem actually wasn't in the code but in the matplotlib backend! The problem clearly occurs when using the
``TkAgg`` backend but does not appear when using the ``Agg`` backend.
- Modified the generation of the QM9 dataset in ``generate_molecule_dataset_from_csv__qm9.py``
- Added the new experiment file ``generate_molecule_dataset_from_csv__qm9sub.py`` which generates the QM9 sub
dataset which is a smaller subset of QM9 with only 22k elements and 9 target columns.
- Added the new experiment ``generate_molecule_dataset_from_csv__aggregators_binary_protonated`` which processes the
larger version of the aggregators dataset where each individual molecule is replaced by all it's protonated variants
- Added the new *background* flavor of visualizing the attributional graph masks. In this method, a filled light green circle
will be painted behind the nodes of the graph.
- Slightly modified the ``ensure_dataset`` function
- Updated the readme file
- Updated the documentation of the standard sub experiments for ``generate_molecule_dataset_from_csv.py``

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

- Fixed a bug where the CogilesEncoder duplicated edges in some very weird edge cases!