dgpy is a generic dependency graph implemented in python, coded live, 10 minutes at a time, in a test-driven development fashion.

You can watch the playlist on youtube and read more about the motivations behind this project at cesarsaez.me.

In terms of features, dgpy implements a null/void node supporting input and output ports/plugs as a base for user specialization, a push and/or pull evaluation model (you can mix it as the model get set per node) and serialization allowing to save and load graphs (reference counting is done during import, so the serialized data is hack-able).

There's no much to document as dgpy doesn't include any builtin node or data type in order to keep it away from any domain specific task, however there's a very simple AddNode implemented in the test suite that could serve as an example (and was used all over the place to drive the development).

Here's a selection of snippets from the test suite showcasing how to get things started.

import dgpy

# Let's implement a simple node
class AddNode(dgpy.VoidNode):
    def initPorts(self):
        super(AddNode, self).initPorts()
        self.addInputPort("value1")
        self.addInputPort("value2")
        self.addOutputPort("result")

    def evaluate(self):
        super(AddNode, self).evaluate()
        result = 0
        for p in self._inputPorts.values():
            if p.value is not None:
                result += p.value
        self.getOutputPort("result").value = result

dgpy.registerNode("AddNode", AddNode)


# Let's create a network
graph = dgpy.Graph()
graph.model = dgpy.PULL

node1 = graph.addNode("node1", AddNode, value1=2, value2=3)
node2 = graph.addNode("node2", AddNode, value1=5)
node2.getInputPort("value2").connect(node1.getOutputPort("result"))

print node2.getOutputPort("result").value  # 5 + 5

node1.getInputPort("value1").value = 10

print node2.getOutputPort("result").value  # 13 + 5


# Let's play with the serialization
data = graph.serialize()
clone = dgpy.Graph.fromData(data)

Check tests.py for more snippets of usage.

This project uses unittest as testing framework (python std library), I'm pretty sure every python developers out there have good reasons to prefer any of the alternatives available but I wanted to keep it simple/accesible to everyone without forcing dependencies.

Coverage at the time this readme was written is 100%, but you can check it by running the test suite.

pip install coverage

coverage run --source=dgpy -m unittest discover
coverage report