sgdBatches: Calculate the default number of batches for a given number of...

In conos: Clustering on Network of Samples

Calculate the default number of batches for a given number of vertices and edges. The formula used is the one used by the 'largeVis' reference implementation. This is substantially less than the recommendation E * 10000 in the original paper.

Description

Calculate the default number of batches for a given number of vertices and edges. The formula used is the one used by the 'largeVis' reference implementation. This is substantially less than the recommendation E * 10000 in the original paper.

Usage

sgdBatches(N, E = 150 * N/2)

Arguments

N	Number of vertices
E	Number of edges (default = 150*N/2)

Value

The recommended number of sgd batches.

Examples

# Observe that increasing K has no effect on processing time
N <- 70000 # MNIST
K <- 10:250
plot(K, sgdBatches(rep(N, length(K)), N * K / 2))

# Observe that processing time scales linarly with N
N <- c(seq(from = 1, to = 10000, by = 100), seq(from = 10000, to = 10000000, by = 1000))
plot(N, sgdBatches(N))

conos documentation built on May 29, 2024, 9:56 a.m.