BPR gradient question

I noticed that the code for calculating gradient in the BPR model computation in the gorse source code is:

`grad := math32.Exp(-diff) / (1.0 + math32.Exp(-diff))`

While in some conventional literature, the formula for calculating gradient seems to be:

$$\frac{1}{1 + e^{-(diff)}}$$

Could you please explain if the author has any special considerations for doing this? What is the reason for this difference?

Start by locating the BPR model computation in the Gorse source and reading the surrounding gradient calculation. Compare the implemented expression with the formula cited in the issue, then clarify whether the difference is intentional. Done would require a documented explanation, but no file, test, or concrete change is identified.

Written by the indexing model from the issue text.