Partitioning Strategies

The Strat class

class Strat(init=True)

The strategy constructor itself calls libscotch’s stratAlloc function, and usually stratInit is called too.

param init:

If True, the graph allocation is followed by the init. If not, the init() method needs to be called before.

init()

This routine Initializes the fields of the Strat object instance. Called by a class constructor when init=True (by default). Mandatory before calling any routine.

exit()

This routine cleans-up all fields of the instance. Is the opposite of the init() routine. Called automatically when the object is destroyed. Calling explicitly this method is only required when an init() call follows.

save(stream)

This routine saves the strategy to the given file or stream in the form of a text string.

param stream:

input file or stream to read from.

type stream:

Either a file object (result of an open(), don’t forget to close it) or a filename, as a string or as a bytes object (such as b"file/name")

graph_cluster_build(flagval, pwgtmax, densmin, bbalval)

This routine provides predefined clustering strategies. The strategy can only be used as a mapping strategy.

param flagval:

The type of clustering strategy to use.

type flagval:

Integer

param pwgtmax:

The maximum cluster vertex weight.

type pwgtmax:

Integer

param densmin:

The minimum edge density.

type densmin:

Float

param bbalval:

The bipartition imbalance ratio.

type bbalval:

Float

graph_map_build(flagval, partnbr, balrat)

This routine fills the strategy with predefined mapping strategies. The strategy can only be used as a mapping strategy.

param flagval:

The type of mapping strategy to use.

type flagval:

Integer

param partnbr:

The number of parts to map to.

type partnbr:

Integer

param balrat:

The bipartition imbalance ratio.

type balrat:

Float

graph_part_ovl_build(flagval, partnbr, balrat)

This routine provides predefined overlap partitioning strategies. The strategy can only be used as a partitioning with overlap strategy.

param flagval:

The type of overlap partitioning strategy to use.

type flagval:

Integer

param partnbr:

The number of parts to partition into.

type partnbr:

Integer

param balrat:

The bipartition imbalance ratio.

type balrat:

Float

graph_order_build(flagval, lvlnbr, balrat)

This routine provides predefined ordering strategies. The strategy can only be used as a graph ordering strategy.

param flagval:

The type of ordering strategy to use.

type flagval:

Integer

param lvlnbr:

The number of levels to use (When the SCOTCH_STRATLEVELMIN or SCOTCH_STRATLEVELMAX or flag is set, the value of lvlnbr is not ignored.)

type lvlnbr:

Integer

param balrat:

The bipartition imbalance ratio.

type balrat:

Float

mesh_order_build(flagval, balrat)

This routine provides predefined ordering strategies. The strategy can only be used as a mesh ordering strategy.

param flagval:

The type of ordering strategy to use.

type flagval:

Integer

param balrat:

The bipartition imbalance ratio.

type balrat:

Float

dgraph_cluster_build(flagval, pwgtmax, densmin, bbalval)

This routine fills the strategy with predefined clustering strategies. The strategy can only be used as a distributed mapping strategy.

param flagval:

The type of clustering strategy to use.

type flagval:

Integer

param pwgtmax:

The maximum cluster vertex weight.

type pwgtmax:

Integer

param densmin:

The minimum edge density.

type densmin:

Float

param bbalval:

The bipartition imbalance ratio.

type bbalval:

Float

dgraph_map(string)

This routine parses the given distributed mapping strategy. The strategy can only be used as a distributed mapping strategy.

param string:

The distributed mapping strategy.

type string:

String

dgraph_map_build(flagval, procnbr, partnbr, balrat)

This routine fills the strategy with the parameters of a default mapping strategy and the given parameters.

param flagval:

The type of distributed mapping strategy to use.

type flagval:

Integer

param procnbr:

The number of processors to map to.

type procnbr:

Integer

param partnbr:

The number of parts to map to.

type partnbr:

Integer

param balrat:

The bipartition imbalance ratio.

type balrat:

Float

dgraph_order(string)

This routine parses the given distributed ordering strategy. The strategy can only be used as a distributed ordering strategy.

param string:

The distributed ordering strategy.

type string:

String

dgraph_order_build(flagval, procnbr, levlnbr, balrat)

This routine fills the strategy with the parameters of a default ordering strategy and the given parameters.

param flagval:

The type of distributed ordering strategy to use.

type flagval:

Integer

param procnbr:

The number of processors to map to.

type procnbr:

Integer

param levlnbr:

The number of levels to use.

graph_bipart(string)

This routine parses the given bipartitioning strategy. The strategy can only be used as a graph bipartitioning strategy.

param string:

The graph bipartitioning strategy.

type string:

String

graph_map(string)

This routine parses the given mapping strategy. The strategy can only be used as a mapping strategy.

param string:

The mapping strategy.

type string:

String

graph_part_ovl(string)

This routine parses the given partitioning strategy. The strategy can only be used as a partitioning with overlap strategy.

param string:

The partitioning strategy.

type string:

String

graph_order(string)

This routine parses the given graph ordering strategy. The strategy can only be used as a graph ordering strategy.

param string:

The graph ordering strategy.

type string:

String

mesh_order(string)

This routine parses the given mesh ordering strategy. The strategy can only be used as a mesh ordering strategy.

param string:

The mesh ordering strategy.

type string:

String

strat_alloc()

This routine ecreates the stratAlloc function, returning a non-initialized Strat instance. It will need the init() method to be called before the instance be built or loaded upon.

graph_part(graph, partnbr, strat, parttab=None)

This routine computes a partition of the graph, with respect to the given strategy.

Parameters:

• graph (Graph) – The graph to partition.

• partnbr (Integer) – The number of parts to partition into.

• strat (Strat) – The partitioning strategy.

• parttab (numpy.ndarray) – The partition array.

graph_part_fixed(graph, partnbr, strat, parttab=None)

This routine computes a partition of the graph, with respect to the given strategy and the fixed vertices in maptab.

Parameters:

• graph (Graph) – The graph to partition.

• partnbr (Integer) – The number of parts to partition into.

• strat (Strat) – The partitioning strategy.

• parttab (numpy.ndarray) – The partition array.

graph_part_ovl(graph, partnbr, strat, parttab=None)

This routine computes a partition with overlap of the graph, with respect to the given strategy and the fixed vertices in maptab.

Parameters:

• graph (Graph) – The graph to partition.

• partnbr (Integer) – The number of parts to partition into.

• strat (Strat) – The partitioning strategy.

• parttab (numpy.ndarray) – The partition array.

Further documentation about these three graph_part* functions can be found in Graph.part(), Graph.part_fixed() and Graph.part_ovl().