Bonus feature: a maximum likelihood estimator that uses #JAX to take derivatives, optimize, and run on the GPU. You give it a likelihood function, it'll return parameter means and covariance matrix. With that, we can implement logit, poisson, and arbitrary variations thereof
-
-
Prikaži ovu nitHvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Wishlist: support for R-style formula syntax, "y ~ x + z" etc

-
Done! Sort of. I just put in a little 3 line parser that is very particular about spaces. Otherwise, I'd have to think about nested parenthesis and LR(1) parsers, and that's when shit gets real
- Još 4 druga odgovora
Novi razgovor -
-
-
Fantastic! Thanks for bringing it to the community
. Do you plan to integrate it with other libs? We now have linearmodels, statsmodels, scikit learn, & your code which do regs - wonder if consolidation would be helpful to users? -
Big fan of statsmodels, and the end goal is generally integration. But wow, how have I never seen linearmodels? They've got most of what I was looking for. Let's see how the benchmarks look.
- Još 4 druga odgovora
Novi razgovor -
-
-
Neat!, Does it support online learning?
-
Well linear.ols is just dead simple OLS, but if you look at http://general.py , it does SGD on maximum likelihood, so you could imagine using that for an online approach
Kraj razgovora
Novi razgovor -
-
-
Benchmarking with reghdfe, fixest etc.?
-
Good question! With reghdfe I'm getting 23s, but that's with absorb so I don't actually get the fe coefficients. Without absorb, memory blows up past 64GB
Kraj razgovora
Novi razgovor -
-
-
Very cool! Do you mind adding an implementation example to the LOST page? https://lost-stats.github.io/Model_Estimation/fixed_effects_in_linear_regression.html …
-
Thanks! I'd be happy to. Will describe the standard approach with Statsmodels then plug fastreg for high-dimensional applications.
- Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.