This vignette contains benchmarks of sgdnet against other similar packages. The data has been precomputed from scripts that are available at https://github.com/jolars/sgdnet/data-raw/.
The benchmarks were generated as follows:
The benchmarks were run on a dedicated Amazon EC2 m4.large instance.
Note that some of the data sets below are not strictly 100% dense, despite the specifications below. They are, however, stored in dense matrix form (the regular
matrix class in R), which makes the packages ignore any sparsity.
In this section, we are going to look at the following datasets:
All of these have been collected from the libsvm binary dataset collection.
For the multinomial model, we have these data sets: