Python API

[mhugo]

Add the ability to write graph plugins in Python. This would involve:

• exposing the plugin API in Python. Classes: Tempus::Plugin, Tempus::Request, Tempus::Roadmap
• exposing the graphs in Python. Tempus::Multimodal::Graph, Tempus::Road::Graph, Tempus::PublicTransport::Graph
• allow the core to load a Python module as a graph plugin

Tasks:

• [x] decide what tool to use for Python bindings. Possibilities: bare cPython, boost.python, sip, swig
• [ ] ~embed an interpreter and allow to run a module~
• [x] write a small (but realistic) example and decide about the exposed API
• [ ] allow the QGIS client to load and use a graph from the API Python (serverless)

[mhugo]

Interesting concept here : https://www.tensorflow.org/versions/0.6.0/get_started/basic_usage.html where the user inputs a "computation graph" that is then "compiled" and send to CPU/GPU. Avoid making actual computations inside Python. Would it make sense for Tempus ?

[mhugo]

Consider the following libs for integration or inspiration:

• https://networkx.github.io/
• https://graph-tool.skewed.de/
• http://igraph.org/redirect.html

[mhugo]

We went for boost python

[mhugo]

We can avoid to embed a Python interpreter by allowing a python module to be loaded as a plugin when the application is made in Python. See for instance an example here https://github.com/Ifsttar/Tempus/blob/dev/python/pytempus/standalone/main.py#L41

The modification of QGIS plugin is postponed to the next release