GraphBuilder

class liesel.model.model.GraphBuilder(to_float32=True)

Bases: object

A graph builder, used to set up a Model.

Constructs a model containing all nodes and variables that were added to the graph builder and their recursive inputs.

Important

• In build_model() , the graph builder will automatically find all inputs to its nodes - and the inputs to these inputs (i.e. it finds inputs recursively).

• The outputs of the nodes, however, are not added to the model automatically, so all root nodes need to be added explicitly.

• Root nodes are nodes that are not inputs to any other node in the graph. The response in a regression model is an example of a root node.

The standard workflow is to create the nodes and variables, add the root var to a graph builder, and construct a model from the graph builder. After the model has been constructed, some methods of the graph builder are not available anymore, because the graph is considered static.

Parameters:

to_float32 (bool) – Whether to convert the dtype of the values of the added nodes from float64 to float32. (default: True)

See also

Model

The liesel model class, representing a static graph.

GraphBuilder.add()

Method for adding variables and nodes to the GraphBuilder.

GraphBuilder.build_model()

Method for building a model from the GraphBuilder.

Var.transform()

Transforms a variable by adding a new transformed variable as an input. This is useful for variables that are constrained to a certain domain, e.g. positive values.

Examples

We start by creating some variables:

>>> a = lsl.Var(1.0, name="a")
>>> b = lsl.Var(2.0, name="b")
>>> c = lsl.Var(lsl.Calc(lambda x, y: x + y, a, b), name="c")

We now initialize a GraphBuilder and add the root node c to it:

>>> gb = lsl.GraphBuilder()
>>> gb.add(c)
GraphBuilder(0 nodes, 1 vars)

We are now ready to build the model:

>>> model = gb.build_model()
>>> model
Model(9 nodes, 3 vars)

Note that when build_model() is called, all weak variables in the graph will be updated. So the value of c is now available:

>>> c.value
3.0

The graph builder is now empty:

>>> gb.vars
[]

Methods

add(*args[, to_float32])	Adds nodes, variables or other graph builders to the graph.
add_groups(*groups[, to_float32])	Adds groups to the graph.
build_model([copy])	Builds a model from the graph.
convert_dtype(from_dtype, to_dtype)	Tries to convert the node values in the graph to the specified data type.
copy()	Returns a shallow copy of the graph builder.
count_node_names()	Counts the number of times each node name occurs in the graph.
count_var_names()	Counts the number of times each variable name occurs in the graph.
groups()	Collects the groups from all nodes and variables.
plot_nodes()	Plots all nodes in the graph.
plot_vars()	Plots all variables in the graph.
rename(pattern, replacement)	Renames all nodes and variables in the graph.
rename_nodes(pattern, replacement)	Renames all nodes in the graph.
rename_vars(pattern, replacement)	Renames all variables in the graph.
replace_node(old, new)	Replaces the old with the new node.
replace_var(old, new)	Replaces the old with the new variable.
transform(var[, bijector])	Transforms a variable by adding a new transformed variable as an input.
update()	Updates all nodes in the graph.

Attributes

log_lik_node	User-defined log-likelihood node, if there is one.
log_prior_node	User-defined log-prior node, if there is one.
log_prob_node	User-defined log-probability node, if there is one.
nodes	The nodes that were explicitly added to the graph.
vars	The variables that were explicitly added to the graph.