TimesFM: Time Series Foundation Model for time-series forecasting

I&#x27;m curious why we seem convinced that this is a task that is possible or something worthy of investigation.<p>I&#x27;ve worked on language models since 2018, even then it was obvious why language was a useful <i>and transferable</i> task. I do not at all feel the same way about general univariate time series that could have any underlying process.

On a related note, Amazon also had a model for time series forecasting called Chronos.<p><a href="https:&#x2F;&#x2F;github.com&#x2F;amazon-science&#x2F;chronos-forecasting">https:&#x2F;&#x2F;github.com&#x2F;amazon-science&#x2F;chronos-forecasting</a>

Seems like a pretty small (low latency) model. Would be interesting to hook up to mouse input (x and y) and see how well it predicts where I’m gonna move the mouse (maybe with and without seeing the predicted path)

&quot;Time series&quot; is such an over-subscribed term. What sorts of time series is this actually useful for?<p>For instance, will it be able to predict dynamics for a machine with thousands of sensors?

&quot;Why would you even try to predict the weather if you know it&#x27;s going to be wrong?&quot;<p>- most OCs on this thread

Blog link (Feb 2024): <a href="https:&#x2F;&#x2F;research.google&#x2F;blog&#x2F;a-decoder-only-foundation-model-for-time-series-forecasting&#x2F;" rel="nofollow">https:&#x2F;&#x2F;research.google&#x2F;blog&#x2F;a-decoder-only-foundation-model...</a><p>Previous discussion: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=39235983">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=39235983</a>

Dear googler or meta-er or timeseries transformer startup something-er: Please make a ChatGPT&#x2F;chat.lmsys.org style interface for one of these that I can throw data at and see what happens.<p>This one looks pretty easy to setup, in fairness, but some other models I&#x27;ve looked at have been surprisingly fiddly &#x2F; locked behind an API.<p>Perhaps such a thing already exists somewhere?

It seems to me that predicting something based on time is rarely accurate and meaningful.<p>Suppose you want to buy stocks? Would you look on a time based graph and buy according to that? Or you rather look at financial data, see earnings, profits? Wouldn&#x27;t a graph that has financial performance on x-axis be more meaningful that one that has time?<p>What if you research real estate in a particular area? Wouldn&#x27;t be square footage a better measure than time?

Is anyone using neural networks for anomaly detection in observability? If so, which model and how many metrics are you supporting per core?

how good is it on stocks?

I&#x27;m not sure I understand two things here. Could someone clarify: 1. This is a foundation model, so you&#x27;re expected to fine tune for your use case, right? (But readme doesn&#x27;t mention tuning) 2. When submitting two series, do they impact each other in predictions?

How can time series model be pre-trained ? I think I’m missing something.

Anyone have insights working with Ikigai’s “Large Graphical Model” and how well it does on time-series? It’s proprietary, but I’m curious how well it performs.

Would this be useful in predicting lat&#x2F;long coordinates along a path? To mitigate issues with GPS drift.<p>If not, what would be a useful model?

is it better than prophet from meta?

If I give this model the first 100 prime numbers, does it give me back the rest of it? If so what is the circuit?

Prophet 2.0

When it comes to time series forecasting, if the method actually works, it sure as hell isn&#x27;t being publicly released.