XGboost, LGBM, pmdarima, stanpy (for bayesian modelling). Plus a few others. Don't ask me what they do with all of these, I'm just the guy who make sure the forecast keeps being reproducible.
Yeah boosted tree models are the shit for tabular data
Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
41–50 of 91 posts
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#42Darts gives you a lot of options, including newer deep learning approaches like NBEATS and NHiTS. https://unit8co.github.io/darts/
+1 for DARTS, a great PyTorch based TS forecasting library that implements several state of the art algos. Devs are responsive on gitter.
A similar pkg is PyTorch-forecasting but overall I prefer DARTS, mainly because of devs responsiveness and better docs
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#43Former Reuters Research Director here. When modeling time series, you will want a model that is sensitive both to short term and longer term movements. In other words, a Long Term Short Term Memory (LSTM). Sepp Hochreiter invented this concept in his Master's thesis supervised by Jürgen Schmidhuber in Munich in the 1990s; today, it's the most-cited type of neural network. Here are papers describing it: https://people…
I don't think LSTMs are state-of-the-art in the domain anymore.
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#44Former Reuters Research Director here. When modeling time series, you will want a model that is sensitive both to short term and longer term movements. In other words, a Long Term Short Term Memory (LSTM). Sepp Hochreiter invented this concept in his Master's thesis supervised by Jürgen Schmidhuber in Munich in the 1990s; today, it's the most-cited type of neural network. Here are papers describing it: https://people…
LSTMs have been going the way of the dinosaurs since 2018. If you really need a complex neural network (over 1D convolution approaches), transformers are the current SOTA. Example implementation in "temporal fusion": https://pytorch-forecasting.readthedocs.io/en/stable/tutoria...
Mind you, in practice I've found these DL approaches overkill for simple problems of the "trends + cyclics + noise" kind.
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#45Forecast package in R is quite useful. Even if you don’t use R, this book by Rob Hyndman is very approachable and easy to follow. https://otexts.com/fpp2/
`fable` is the successor to `forecast`, according to Hyndman, though the latter is still maintained.
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#46Stuff I've used: - Prophet - seems to be the current 'standard' choice - ARIMA - Classical choice - Exponential Moving Average - dead simple to implement, works well for stuff that's a time series but not very seasonal - Kalman/Statespace model - used by Splunk's predict[1] command (pretty sure I always used LLP5) I did some anomaly detection work, in business transactions, and found the best way was to create a sort…
I don't know what you've been using Prophet for but I found it to be very brittle.
Yup, as per my other comment (https://news.ycombinator.com/context?id=33448802), fbprophet is largely tuned for a few years of somewhat regular business data sampled daily (e.g. sales per day). Outside its comfort zone (e.g. if you have monthly, or hourly/minutely data, or step changes / level shifts) it can fall apart pretty quickly. But its comfort zone happens to be very popular in business settings.
To be fair, it's pretty hard to create generic models that are can robustly handle any random time series.
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#47Everything is either light gbm or an early exploratory experiment.
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#48Stuff I've used: - Prophet - seems to be the current 'standard' choice - ARIMA - Classical choice - Exponential Moving Average - dead simple to implement, works well for stuff that's a time series but not very seasonal - Kalman/Statespace model - used by Splunk's predict[1] command (pretty sure I always used LLP5) I did some anomaly detection work, in business transactions, and found the best way was to create a sort…
> - 2-3 anomalies in a row
> - high deviation from expected
> - multiple models all detected anomaly
This is basically what statistical process control charts do for you. If you haven't learned about it already, I can recommend looking it up!
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#49Earlier quoted context omitted.
I don't know what you've been using Prophet for but I found it to be very brittle.
Yup, as per my other comment ( https://news.ycombinator.com/context?id=33448802 ), fbprophet is largely tuned for a few years of somewhat regular business data sampled daily (e.g. sales per day). Outside its comfort zone (e.g. if you have monthly, or hourly/minutely data, or step changes / level shifts) it can fall apart pretty quickly. But its comfort zone happens to be very popular in business settings. To be fair,…
Fair enough, when I was working with monthly sales, it was pretty bad.
Re: Ask HN: Data Scientists, what libraries do you use for timeseries forecasting?
#50I really like FB's Prophet: https://facebook.github.io/prophet/