[Request] Force Warning for Non-convergence about fixest

[lrberge]

Hi Luke, that's a very fair request!

Out of curiosity: do you happen to have variables with varying slopes in your estimation?

It's because the only situations I experienced in which the algorithm hit the iteration limit was when there were several variables with varying slopes. It's a known issue with a known fix (I still have to implement it though). Currently there's a warning in case of convergence problems but only in those situations.

As suggested, I'll add a warning in case of non-convergence and will also increase the number of iterations by default. (Actually I didn't implement it because I thought--more precisely hoped [no doubt too strongly]--that 2000 iterations would always be enough.)

[clukewatson]

Yes, I did (or at least this was the case that I noticed the issue).

Awesome. Do you want me to close the issue, or I can leave that up to you?

[lrberge]

Thanks, I'll close it once I've made the changes in the code. In the meantime it serves me as a reminder.

[lrberge]

I've added a warning and the message will also pop in the print to really signal the problem.
Since there's not point to be fast if the results are wrong :-D, I've also increased the default number of iterations to 10,000.
Thanks again!