Tool to explore regularly sampled time series

If you’ve ever wrestled a noisy, regularly sampled time series into submission, tseda might feel like a small mercy. The new open-source app, published on GitHub and discussed on Hacker News, is built for exploratory analysis of series sampled at hourly cadence or slower. Short answer: it bundles quick distribution checks, autocorrelation diagnostics, Singular Spectral Analysis (SSA) decomposition, change‑point detection and an observation logging page — all in a lightweight Dash app and a companion notebook. Clean, fast, and a little nerdy. Exactly what some analysts need.

What it does

Tseda guides users through a three‑step workflow: an initial assessment (kernel density, box plot, PACF/ACF), SSA‑based decomposition (heuristic window sizes — e.g., hourly→24, monthly→12, quarterly→4 — which you can tweak in the UI), and observation logging that includes AIC as a function of model rank. Reconstruction plots, explained‑variance diagnostics and a change‑point view complete the toolkit. It has been reported that tseda currently limits uploads to 2,000 samples by default (this cap is configurable). Want to peek under the hood? There’s also a notebook interface and example data loaders for one‑off analyses.

Getting started

You’ll need Python 3.13 or newer. Conda is the recommended route for development (conda create -n tseda python=3.13; pip install tseda), but there’s a non‑developer quick start via pipx (pipx install tseda; tseda). The app runs locally by default at http://127.0.0.1:8050 and accepts CSV/Excel files with a timestamp column first and a numeric column second. Files must be regularly sampled (hourly or coarser) and contain no missing values — clean data only, please.

Caveats and context

Tseda isn’t a black‑box forecasting engine. It’s an exploratory swiss army knife: help for pattern spotting, seasonality guessing and preliminary model selection, not a production forecasting pipeline. That said, for anyone tired of ad hoc plotting — and for those who like SSA’s blend of linear algebra and signal smoothing — it’s a practical, opinionated tool that could speed up the early stages of analysis. Want less hand‑wringing and more insight? Give it a spin and see if it sticks.

Sources: github.com/rajivsam, Hacker News