Nixtla转发了
As a time series enthusiast, I couldn’t resist giving Nicolas Vandeput's competition a shot (although I hadn’t planned on participating due to my personal holiday plans). After trying a few statistical models, I took my laptop to a coffee shop this morning and decided to try some tree-based models. Two coffees later, I built a model that now has me in 9th place (around the top 5% of all participants). Huge shout-out to Nixtla for building an amazing library that made it easy to experiment with different models, feature engineering techniques and quickly evaluate performance. If you're into time series forecasting, I highly recommend checking it out—it's a game changer! Most likely, I won’t have time to participate further, but I wish the best of luck to all participants. I can't wait to dive deeper into the solutions once the competition wraps up, and I'm looking forward to sharing my notebook soon. #TimeSeries #DataScience #MachineLearning #Forecasting #Nixtla #AI #FeatureEngineering
Jakub Figura congrats! I am glad to read that I am not the only one to enjoy deep work on models :)
Well done. I’ve got the error down to 0.529 using traditional stats models but I can’t get my tree or neural models below that. Will you share code once competition is over?
Hey Jakub, fascinated by your post. Just wanna ask are you forecasting multiple horizon recursively, or are you doing something like a model with multiple outputs at once?
Data Science | ITAM | Power Markets
5 个月Daniel Vanden Brink tree based model