Project

You can run ZnTrack nodes in three different ways. Let’s assume the following two nodes:

import zntrack

class AddOne(zntrack.Node):
    parameter: int = zntrack.zn.params()
    outputs: int = zntrack.zn.outputs()

    def run(self):
        self.outputs = self.parameter + 1

class SubtractOne(zntrack.Node):
    parameter: int = zntrack.zn.deps()
    outputs: int = zntrack.zn.outputs()

    def run(self):
        self.outputs = self.parameter - 1

Direct Use

ZnTrack Nodes are designed to be used as regular Python classes. This allows for easy debugging and testing. Typically, you would use Eager graph or DVC for production.

node1 = AddOne(parameter=1)
node1.run()

node2 = SubtractOne(parameter=node1.outputs)
node2.run()

assert node1.outputs == 2
assert node2.outputs == 1

Eager graph

You can use ZnTrack without DVC. This closely resembles the ZnFlow package. ZnTrack is built on top of it. The general API would look as follows:

with zntrack.Project() as project:
    node1 = AddOne(parameter=1)
    node2 = SubtractOne(parameter=node1.outputs)

project.run(eager=True)
assert node1.outputs == 2
assert node2.outputs == 1

DVC

The main purpose of ZnTrack is to use it with DVC. If you use project.run() it will serialize the Nodes inside the project and run them with DVC. You can use project.run(repro=False) to only build the graph and execute it later.

with zntrack.Project() as project:
    node1 = AddOne(parameter=1)
    node2 = SubtractOne(parameter=node1.outputs)

project.run()

node1.load()
node2.load()

assert node1.outputs == 2
assert node2.outputs == 1