Cristian Arteaga

Fantastic! Thank you so much for suggesting those resources and for offering your help. I just emailed you.

Hi @jhelvy. I integrated your probability and gradient formulation and the estimation is now around **~40% faster.** It is impressive how your formulation significantly reduces the amount of computations. Nice...

Yes, I am quite happy with the 40% speed gains. Thank you for sharing your valuable formulation to make it faster. In xlogit, I do not iterate over individuals, betas,...

Yes! In Python, I use Sphinx to automatically generate the documentation. Also, I use Jupyter notebooks to create examples, and I further convert these to web page using Sphinx. For...

Currently, the main page has the following links (see screenshot below) to the documentation website, but I agree that it is a good idea to additionally create an "About" section...

Oh I see. Yes, it is a great idea. I just added it. Thanks for the suggestion.

Hi @jhelvy, I managed to get the panel data part to work. Now I am having a little challenge estimating fixed (non-random) coefficients, so I figured I would ask in...

I see. Does this mean that you use the same gradient computations for fixed and random coefficients with the only difference being that you set the random draws as zero...

Yes, I am actually surprised too because the math works and makes total sense, so I also suspect it's some weird bug. Yes, by looking at the gradients, I can...

I finally figured it out. The issue was because, while computing the gradient, I was doing the operation $V_{nj} - V_{n*}$ separately for fixed and random coefficients. However, such an...