awesome speech, thanks! can't believe noone asked about model selection or current sota's been using boostings for forecasting for like 4.5 years already and still can't be beaten as an optimal prod solution. What are you using mostly as a compromise between quality and speed (+scalability), what about self-attention and CN-based attention nets? let aside interpretability
Really interesting library. I have questions which maybe someone can answer here: 1) Can we use our custom model like neural network or LSTM? and 2) How this library assure the time series is exchangeability for the conformal prediction?
Awesome talk!
awesome speech, thanks!
can't believe noone asked about model selection or current sota's
been using boostings for forecasting for like 4.5 years already and still can't be beaten as an optimal prod solution. What are you using mostly as a compromise between quality and speed (+scalability), what about self-attention and CN-based attention nets? let aside interpretability
Really interesting library. I have questions which maybe someone can answer here: 1) Can we use our custom model like neural network or LSTM? and 2) How this library assure the time series is exchangeability for the conformal prediction?
Nixtla
How can we provide confidence intervals for classic statistical models that are not stochastic?
Is there a video for anomaly detection using Nixtla?
Great talk, tons of thanks