[HN Gopher] Merlion: A Machine Learning Framework for Time Serie...
___________________________________________________________________
Merlion: A Machine Learning Framework for Time Series Intelligence
Author : klaussilveira
Score : 77 points
Date : 2025-02-28 18:59 UTC (4 hours ago)
(HTM) web link (github.com)
(TXT) w3m dump (github.com)
| guskel wrote:
| The "Comparison with Related Libraries" section is notably
| missing comparison against aeon and sktime, which also aim to be
| the kind of holistic time-series library like Merlion.
| lamename wrote:
| I don't see `tsai` in there either
| jldugger wrote:
| I do wish these were easier to integrate with prometheus or
| graphite. They both have rudimentary forecasting models but both
| leave much to be desired in parameterization.
| esafak wrote:
| Same here. The open source space here is neglected. I just
| heard about Grafana's augurs: https://github.com/grafana/augurs
| xnx wrote:
| Is this a competitor to Google's TimeFM?
| https://github.com/google-research/timesfm
| macleginn wrote:
| TimeFM is a single pre-trained decoder-only model for time-
| series forecasting. Merlion is a collection of models, both
| neural and more traditional, for a wide variety of time-series-
| related tasks. I can imagine TimeFM ending up in the set
| eventually, given right circumstances.
| xnx wrote:
| Thanks for the summary explainer. Very hard to keep up with
| the abundance of models if you don't work directly in that
| area.
| ValentineC wrote:
| Hopefully Salesforce Legal cleared everything surrounding this
| project's name and logo:
|
| https://www.stb.gov.sg/about-stb/stb-brand-assets#94e2cb04e4...
|
| That said, it also looks like this project has been around for 3+
| years without issue.
| theanonymousone wrote:
| A warm welcome to the new buzzword expression "Time series
| intelligence", further muddying the water of AI terminology and
| confusing everyone.
| janalsncm wrote:
| Time series is an established term. For example predicting an
| earthquake from a sequence of events, or a stock price change,
| etc.
|
| I never worked very seriously with time series problems, but I
| know enough from the title to know I can't evaluate how good
| this thing is.
___________________________________________________________________
(page generated 2025-02-28 23:00 UTC)