This message was deleted.

From my perspective the authors make bad claims. The title of the paper literally states it’s for “at scale” forecasting but takes like a second to fit for one time series and doesn’t have many speed ups available. It doesn’t really do well in most scenarios. It Leaves wayyy too much signal in the residuals. Hard Changepoints are also really scary to productionalize without knowing what you are doing but none of these points are ever brought up by the authors, instead they mostly say (at least used to) that it was a good automatic solution for time series. Now they have changed their tune but the damage is done. Prophet can win but so can forecasting the last value plus 10, doesn’t mean it’s actually good.

If you have time series with a clear trend and clear multiple seasonalities, it usually works quite well, often outperforming other approaches like ARIMA+MSTL in my experience. Additionally, you don't always find useful information in the residuals. Ignoring them can sometimes enhance robustness to noise, resulting in a simpler model that is easier to explain and interpret. However, you can also combine Prophet with another model to address residuals, or use NeuralProphet, which can also leverage any remaining residuals after modeling the trend, seasonalities, and exogenous variables.

I don’t think ARIMA or MSTL are super fantastic either. If you look at benchmarks like from monash or Peter Cotton’s micropredictions prophet lags miles behind. I probably don’t hate it as much as others but it’s not worth the time to fit it. If the trend is clear and you have multiple seasonality then it’s way easier to fit your own procedure or you can fit numerous other better optimized methods in the time it takes to fit prophet. From a professional note when we used to have prophet it would account for a very small percentage of “wins” in our stat engine.

Honestly, the benchmarks by Peter Cotton are more or less garbage. Prophet performs well with specific types of time series, like the one I mentioned earlier, which aren't included in Peter Cotton's benchmarks. If you have a time series without a clear trend or clear multiple seasonalities, and you can forecast it in an autoregressive way, it's obvious that Prophet won't work well for it, making such benchmarking entirely pointless. If you have mastered forecasting algorithms and understand when to use them based on their peculiarities, having a library like Prophet in your toolbox is still useful. P.S. I haven't used Prophet in a while, but according to the Git history, it seems that the latest versions have been significantly sped up. Additionally, Prophet can be easily parallelized using Spark or other distributed computing platforms.

last I tried pip installing it failed haha. But yeah Micropredictions is not super useful but is still a data point. We also have M4 and M5 both of which does not have a good showing for Prophet, I am not aware of a large benchmark that does show well for Prophet. Most of the time I see prophet showing up in like wind/solar predictions and other fields that are newer to time series. At the end of the day - there are just way better alternatives....like some of my work 😀 But ultimately that is why people don't like it - large promises with not much delivery on those promises and an overabundance of its use due to the parent company attached to it.

@Manuel what’s your Twitter handle?

It is subjective, I have read a lot of his articles and in my opinion he is not worth much. I could easily confront him and dismantle many of his statements without any problem.

@Manuel just curious what statements you are referring to? He is in a vastly different field than me but he is still pretty well respected from all I’ve seen, gives talks at conferences like the M5 conference.

For example, many articles on www.microprediction.com precisely about Facebook's Prophet library, where he joins the bandwagon of critics without acknowledging that Prophet can be useful in certain use cases. The main issue with Prophet is that it was perhaps initially oversold as a universal solution for all forecasting problems and this was partly due to the limited availability of forecasting libraries in Python at the time of Prophet's introduction. Prophet is actually well-suited only for specific types of time series. Criticizing Prophet indiscriminately, and trying to demonstrate its ineffectiveness with time series data for which its algorithm is clearly not designed, is just plain stupid. It's akin to using a fishing pole and complaining that it can't be used to catch wild boar. It may have been mistakenly believed that it could do that, however it's unfair to say that a fishing pole isn't useful when the objective is to catch fish, not wild boar.

@Manuel I think his critique is spot on: https://www.microprediction.com/blog/prophet. He even does a psuedo-meta analysis and comes to the same conclusions I did-no real benchmarks show favorably for Prophet except small one-off examples with small samples size. The problem is that it is 'designed' for deterministic trends with changepoints and deterministic seasonality which is like no real-world examples. The algorithm has no sense of 'level' meaning it can over/under forecast for a long period of time, this makes it ill-suited for most cases as it will start its forecast way off. So the time series it is suited for are already super easy to forecast for and even then other methods will outperform it. I understand liking Prophet, the forecasts make sense to us unlike other more complicated approaches. It just draws lines! But just like in the ML world, the simpler methods that a toddler can understand are just outclassed in the real world. If Prophet performs best for your data that's great, publish the results comparing boosted trees, smoothers, autoarima, TBATS, nbeats, nhits and everything else out there. But, if you are still leaning on 'sometimes it does better than ARIMA' just remember that EVERY method can get lucky.

@Manuel do so, please do what you stated you can do https://nixtlacommunity.slack.com/archives/C031EJJMH46/p1701450178794499?thread_ts=1700146613.954629&channel=C031EJJMH46&message_ts=1701450178.794499

@Valeriy I simply made an observation about a comment I read in a post linked in this channel, it was a comment by a forecasting practitioner in a channel of forecasting practitioners, nothing more. I did not even mention Peter first but only responded to Tyler who mentioned his benchmarks. However, I saw that you and Peter are friends and he even tagged you on some of his posts, however, I did not know that you were also his personal defense attorney.

What post was linked in this slack? This slack is for Nixtla library and I don’t think you slandering people in this slack is either acceptable or relevant to this slack. I am not friends with Peter as you continue to falsely claim and I don’t understand why you are doing this. Tagging @Max (Nixtla) here to show what goes on in this channel.

The poll about forecasting libraries

There was nothing in this poll about facebook prophet.

It was a comment in the linked poll

What comment?

It's a comment in the comment section of the poll linked in this channel

And? This gives you excuse to slander people in this slack? I am pretty sure this is not what this slack is for.

I did not slander anyone, I simply responded, arguing extensively, to Tyler's messages. That is until the intervention of his personal lawyer who, without responding on the merits of the conversation, basically told me that his friend is the best in the world and is never wrong.

If your comments are not slander I don’t know what else is.

Exactly, instead of intervening in the merits of the discussion you intervened only to say that your friend is the best in the world. Give me an honest answer: do you think you wrote an intelligent message?

This becomes really interesting, so you think Nixtla allows people to slander others in this slack if they are “contributors” to stats forecast?

You still haven't told me whether you think you wrote an intelligent message

Hi @Valeriy, @Manuel and @Tyler Blume, we really appreciate the passion and energy of this discussion. However, it seems that the main points have been made and we can all agree to disagree. This discussion is not leading anywhere and I think it would be best if we just stopped this thread. We really appreciate the contributions that each of you have made to this community.