https://github.com/nixtla logo
m

Max (Nixtla)

04/28/2022, 3:06 PM
@Daniel Falbel, welcome!
d

Daniel Falbel

04/28/2022, 3:29 PM
Hi @Max (Nixtla)! Nice to meet you! Here mostly for learning from you 🙂 I am currently working on implementing neural time series models in R. I started with TFT: https://github.com/mlverse/tft
m

Max (Nixtla)

04/28/2022, 4:11 PM
Thats great! Happy to help in any way.
@Kin Gtz. Olivares wants to implement TFT in Neuralforecast too.
👍 2
k

Kin Gtz. Olivares

04/28/2022, 6:59 PM
I was taking first steps reading the pytorch-forecasting version
Their dataloader and dependencies were not too clear
@Daniel Falbel have you validated the mlverse/tft?
d

Daniel Falbel

04/28/2022, 7:39 PM
For mlverse/tft I mostly followed the paper + pytorch-forecasting code and checked a few details on the original implementation. Sorry @Kin Gtz. Olivares what do you mean by validating? I’m trying to reproduce the paper benchmarks, but it’s hard to tell what exactly they report, but results are making sense.
k

Kin Gtz. Olivares

04/28/2022, 7:47 PM
sounds good, promising results