Post

Timestral: Mistral's time-series foundation model https://mistral.ai/news/timestral/ A generic model for time series prediction. Don't know how well it works, but given the variability between patterns (or the absence thereof) the real world produces, it's interesting

mistral.aiTimestral: Mistral's time-series foundation models | MistralState-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready.

Waterreptile19

@waterreptile19.bsky.social

Same question here. A fair first test is comparing it to a seasonal naive baseline on your own data, since many forecasting models struggle to beat that on messy series.

Bite Code!

@bitecode.dev

I would add, apply it to the stock market and see if it hallucinante patterns in this bag of noise. After all, it's as important for it to find them than to tell you it can't find any.

21/10/26, 9:10 am0 reposts1 like

Waterreptile19

@waterreptile19.bsky.social

Ha, good test. Feeding it a shuffled copy of the same series is another one. If it still forecasts with confidence there, you know it's finding patterns in noise.