past.Utils.Ripplewalk_sampler

past.Utils.Ripplewalk_sampler(graph, r=0.5, batchsize=6400, total_times=10)

Training stategy of subgraph segmentation based on random walk, enabling mini-batch training on large datasets

Parameters:

  • graph – graph indicating connectivity between spots, usually K-NN graph constructed from spatial coordinates

  • r – expansion ratio for sampling subgraph

  • batchsize – number of samples for a mini-batch

  • total_times – decide the number of subgraph to sample, number of subgraph = total_times * dataset_size / batchsize

Returns:

a list containing index of sampled sub-graphs

Return type:

list