nlgm.optimizers Module

class nlgm.optimizers.RandomWalkOptimizer(graph, signatures, start, criterion)

Random-walk optimizer over a graph of candidate signatures.

Parameters:

graphnumpy.ndarray

Square adjacency matrix.

signatureslist

Signature values indexed by graph node.

startint

Starting node index.

criterionstr

Optimization criterion label.

Methods

normalize_graph()	Normalize transition weights row-wise.
optimize(objective, max_iters[, callback])	Optimize objective with random-neighbor traversal.
optimize_with_backtracking(objective, max_iters)	Optimize objective using random walk with path backtracking.

normalize_graph()

Normalize transition weights row-wise.

optimize(objective, max_iters, callback=None)

Optimize objective with random-neighbor traversal.

Parameters:

objectivecallable

Function taking a signature and returning (metric, loss).

max_itersint

Maximum number of walk iterations.

callbackcallable, optional

Callback invoked as callback(signature, (metric, loss)).

Returns:

tuple

Best result as (signature, metric, loss).

optimize_with_backtracking(objective, max_iters, callback=None)

Optimize objective using random walk with path backtracking.

Parameters:

objectivecallable

Function taking a signature and returning (metric, loss).

max_itersint

Maximum number of walk iterations.

callbackcallable, optional

Callback invoked as callback(signature, (metric, loss)).

Returns:

tuple

Best result as (signature, metric, loss).