.xoptim.order_optim

class experimental_experiment.xoptim.order_optim.OrderAlgorithm(*values)

Defines the possible order algorithm.

• NONE: does not change anything

• RANDOM: random order

• SHAPE: moves every shape node just behind the node producing its input

class experimental_experiment.xoptim.order_optim.OrderOptimization(builder: GraphBuilder, algorithm: OrderAlgorithm = OrderAlgorithm.NONE, verbose: int = 0)

Optimizes the order of computation. It tries to minimize the distance between a producer and the consumer or a results. The idea is to reduce the memory usage.

Parameters:

• builder – GraphBuilder holding the model

• algorithm – to apply

• verbose – verbosity

optimize() → List[Dict[str, Any]]

Optimizes the model inplace. It optimizes the model in the builder itself by switching nodes.

random_order()

Moves nodes by random.

shape_order()

Moves shape right after the node it consumes is created.