Andrew Landgraf
Andrew Landgraf
It seems to be okay with `k >= 2`
Also seems okay with `main_effects = FALSE` and `k = 1`. I'm pretty sure the main effects are the problem.
I haven't worked with libffm before, but the API is pretty similar to libfm. If I ever have an excuse to use libffm, I will try it out. But for...
My only suggestion is to make sure you have rARPACK and set partial_decomp = TRUE. On Sun, Jul 25, 2021 at 8:18 PM Oliver Stuart ***@***.***> wrote: > Hi, >...
I looked into this and it looks like each row requires a matrix inversion. This may be to computationally prohibitive for this package. Can look into again later.
I agree they are largely similar, but an advantage of SGNS is that it does better for rarely occurring words. As the [Swivel](https://arxiv.org/pdf/1602.02215.pdf) paper puts it: "GloVe is under-constrained: there...
That is a downside. However, my intuition is that the number of parameter updates is more relevant than than number of epochs. For example, Figure 5 of the [BPR paper](https://arxiv.org/ftp/arxiv/papers/1205/1205.2618.pdf)....