RCC, FastAssign, and SwapEdges

[GomezGab]

LAGraph_RichClubCoefficient.c, rcc_demo.c, test_RichClubCoefficient.c, and LG_check_RCC.c

• Get the rich club coefficient of a graph.
• Given a Symmetric Graph with no self-edges, LAGraph_RichClubCoefficient will calculate the rich club coefficients of the graph.
• The values will be output as a sparse GrB_Vector, and the rich club coefficient of k will be found at the closest entry at or above k.
• check_RCC is the pure C version of the algorithm

LAGraph_FastAssign.c and FastAssign_demo.c

• A GBv10 algorithm
• Essentially moves CSC trick found in FastSV7 and in SwapEdges (WIP) to its own function.
• This method is used for cases where you want to build a vector equivalent to the following for loop:

  for (j = 0 ; j < n ; j++)
    {
        uint64_t i = I_vec [j] ;
        c [i] += X_vec [j] ;
    }
  

• Where += is defined by the accum and dup operators.
• See more details in function comments.

LG_CC_FastSV7_FA.c, cc_demo.c, and test_ConnectedComponents.c

• Demonstrates a use case for FastAssign by using FastAssign in the FastSV7 algorithm for GraphBLASv10.
• Passes LG_BRUTAL tests
• Code is simpler but performs as well as FastSV7. (In fact, it is consistently ~5% faster on GAP-twitter in my testing)
• Piggybacked off of cc_demo.c and test_ConnectedComponents.c for testing and benchmarking.

LG_CC_FastSV7.c and LAGraph.h

• Moved out the definition for USING_GRAPHBLAS_V10

LAGr_SwapEdges.c, LAGraph_SwapEdges.c, test_SwapEdges.c, and SwapEdges_demo.c

• Randomizes undirected, unweighted Graphs.
• LAGr allows the user to customize attempted swaps per edge, minimum swap count per edge, and the number of swaps to do.
• LAGraph_SwapEdges only allows the user to provide the number of desired swaps pre-edge to perform on a Graph, with the other values predetermined.

[GomezGab]

I defined LAGRAPH_INSUFFICIENT_SWAPS as 2100 in LAGraphX.h. Was this correct?
Should I return the number of swaps I made? As its own output variable, or do I make totSwaps input/output?

[DrTimothyAldenDavis]

I'm reluctant to add a new enum value to the LAGraph return values that is unique to just one algorithm. How about returning LAGRAPH_CONVERGENCE_FAILURE instead? The method iterates, trying to find swaps, and if it doesn't succeed, it does need to return an error.

The description of LAGRAPH_CONVERGENCE_FAILURE is quite similar, "An iterative process failed to converge to a good solution".

What do you think?

[GomezGab]

I think returning an error is reasonable. Although currently, I return the error, but still return a valid G_new graph with however many swaps I managed to get before going under the per-loop threshold. Is this a reasonable thing to do, or does returning an error imply the output graph should be freed?

[DrTimothyAldenDavis]

I see. Maybe we should talk it over. In your current design, LAGRAPH_INSUFFICIENT_SWAPS is a warning, not an error, right?

[DrTimothyAldenDavis]

Since this #define is internal, but accessible to any LAGraph algorithm, let's rename it to "LG_SUITESPARSE_GRAPHBLAS_V10". That way, it mimics the LAGRAPH_SUITESPARSE definition (which is in LAGraph.h), but it has an LG_* prefix which means it's for internal use only.

[DrTimothyAldenDavis]

Is this ready to merge?

[GomezGab]

Ready to go

[DrTimothyAldenDavis]

Looks great. Merged into the with_GraphBLAS_v10 branch.