April 13, 2026
Graphs, gaffes, and 8050 laughs
Tool to explore regularly sampled time series
New time‑series toy drops — fans cheer, skeptics roast the limits
TLDR: Tseda is a new click‑to‑explore tool for hourly/daily time series that splits out trend and seasonality and runs locally, but requires Python 3.13. The community is split between praising the easy workflow and mocking the 2,000‑row cap, clean‑only data rules, and “old‑school” methods—yet many still want to try it.
A new point‑and‑click app called tseda promises to make time series (think hourly, daily, monthly measurements over time) way less scary: upload a CSV, peek at patterns with correlation charts (ACF/PACF), split your data into trend vs cycles using SSA (a way to separate signal from noise), spot change points, and get an automatic summary—then run it locally at http://127.0.0.1:8050. Sounds dreamy… until the comments rolled in.
The crowd instantly split. The “finally, no more notebook spaghetti” camp loves the one‑command launch (pipx/conda), the clean three‑step flow, and the fact you can tweak the seasonal window. They say it’s perfect for quick sanity checks and demoing to non‑coders. But the real‑world warriors are roasting the rules: only regularly spaced data, no missing values, and a default 2,000‑row cap. Cue the memes: “2000 rows? My smartwatch collects more before lunch,” and “In real data, missing values ARE the seasonality.”
Then came the version drama: requiring Python 3.13 lit up enterprise folks stuck on older setups, while devs clap back that modern‑only keeps things stable. Purists bickered over methods: “ACF/PACF are grandpa charts” vs “they still slap,” and “SSA beats Prophet/ARIMA for exploration” vs “just use something standard.” Dash running on port 8050? Bingo meme achieved. Even conda vs pip skirmishes broke out.
Love it or roast it, everyone agrees: if your data plays by the rules, tseda makes eyeballing patterns fast—and the debate is half the fun.
Key Points
- •tseda is a Python-based tool for exploring regularly sampled time series (hourly or lower frequency), with a 2,000-sample default limit.
- •The workflow includes initial assessment (KDE, box plot, ACF, PACF), SSA-based decomposition with adjustable window sizes, and observation logging.
- •SSA windows are heuristically set by sampling frequency (hourly=24, monthly=12, quarterly=4) and can be changed in the UI.
- •Observation logging uses AIC for model selection by plotting AIC versus model rank and provides an automatic summary.
- •Installation requires Python 3.13+ and supports conda (recommended), pipx quick start, or pip; the Dash app runs locally at http://127.0.0.1:8050 with host/port overrides.