Create a model

The model is built based on the graph and the desired handler. A mmodel handler is an execution method that handles the node execution order and intermediate value flow. The resulting instance is a callable that behaves like a function. The resulting model copies and freezes the graph, making it immutable. A handler class is passed to the handler argument, and any additional arguments of the handler can be passed as a dictionary to the handler_kwargs argument.

First, we define the nodes and the graph.

import numpy as np
import math
from mmodel import Graph, Node

grouped_edges = [
    ("add", ["log", "func node"]),
    ("log", "func node"),
]

def func(sum_xy, log_xy):
    """Function that adds a value to the multiplied inputs."""

    return sum_xy * log_xy + 6

# define note objects
node_objects = [
    Node("add", np.add, inputs=["x", "y"], output="sum_xy"),
    Node("log", math.log, inputs=["sum_xy", "log_base"], output="log_xy"),
    Node("func node", func, output="result"),
]

G = Graph(name="example_graph")
G.add_grouped_edges_from(grouped_edges)
G.set_node_objects_from(node_objects)

Then, we create the model instance.

from mmodel import MemHandler

model = Model(name='model', graph=G, handler=MemHandler, doc="Test model.")

>>> model
<mmodel.model.Model 'model'>

>>> print(model)
model(log_base, x, y)
returns: result
graph: example_graph
handler: MemHandler

Test model.

# visualize the graph
>>> model.visualize()

Note

The graph cannot have cycles.

The model determines the parameter for the model instance.

edit the model

The model can be edited by appling one or multiple change of the arguments. For example if we want to change the documentation of the model, we can use the edit method. A new model instance is returned.

new_model = model.edit(doc="New documentation.")

>>> print(new_model)
model(log_base, x, y)
returns: result
graph: example_graph
handler: MemHandler

New documentation.

extra return variables

The default return of the model is the output of the terminal nodes. To output intermediate variables, use “returns” to define additional output for the model. All other handler parameters can be directly passed as keyword arguments.

See handler reference for all available handlers.

apply defaults to parameters

Any parameter default values should be applied at the model level. To add default values, use the “param_defaults” argument:

model = Model(name='model', graph=G, handler=MemHandler, doc="Test model.",
              param_defaults={"log_base": 10})

>>> print(model)
model(x, y, log_base=10)
returns: result
graph: example_graph
handler: MemHandler

Test model.