Create a model group

In the mmodel design, we intend to keep the graph behavior similar to the NetworkX graph. Therefore, all the nodes and edges are used in the final graph. However, there are cases that graphs and models shares the similar pool of nodes but different edges.

In this case, a model group can be created to store multiple similar graphs and models in a single object. The parameters used in the ModelGroup class are:

1. “name”, string: model group name

2. “node_objects”, list: list of node objects

3. “model_recipes”, dict: dictionary of model recipes, the keys are the model names

4. “model_defaults”, dict: shared model parameters between different models

5. “doc”, string: model group documentation

from mmodel import Node, ModelGroup, MemHandler
import numpy as np
import math

# 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("sub", np.subtract, inputs=["sum_xy", "log_xy"], output="sub_xy"),
    Node("mul", np.multiply, inputs=["sum_xy", "log_xy"], output="mul_xy"),
]

# create graph edges
grouped_edges_A = [
    ("add", ["log", "sub"]),
    ("log", "sub"),
]

grouped_edges_B = [
    ("add", ["log", "mul"]),
    ("log", "mul"),
]

model_recipes = {
    "model_A": {"grouped_edges": grouped_edges_A},
    "model_B": {"grouped_edges": grouped_edges_B},
}

model_defaults = {"handler": MemHandler}
model_group = ModelGroup(
    "example_model_group",
    node_objects,
    model_recipes,
    model_defaults,
    doc="Test model group.",
)

>>> model_group
<mmodel.model.ModelGroup 'example_model_group'>

>>> print(model_group)
example_model_group
models: ['model_A', 'model_B']
nodes: ['add', 'log', 'sub', 'mul']
model_defaults:
- handler: <class 'mmodel.handler.MemHandler'>

Test model group.

The models can be extracted and executed.

model_A = model_group.models["model_A"]
model_B = model_group.models["model_B"]

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

>>> print(model_A)
model_A(log_base, x, y)
returns: sub_xy
group: example_model_group
graph: model_A_graph
handler: MemHandler

>>> model_A(2, 5, 3)
5.0

>>> model_B(2, 3, 4)
24.0

edit the model group

The model group can be edited by applying one or multiple changes to the arguments. A new model group instance is returned.

new_model_group = model_group.edit(doc="New documentation.")

>>> print(new_model_group)
example_model_group
models: ['model_A', 'model_B']
nodes: ['add', 'log', 'sub', 'mul']
model_defaults:
- handler: <class 'mmodel.handler.MemHandler'>

New documentation.