Univariate workflows

When to use univariate methods

Use the univariate path when one observed series is decomposed into trend, seasonality, and remainder components.

Good starting methods:

• STL when the seasonal period is known and the data are reasonably regular,
• SSA when you want a flexible subspace method,
• STD when you want blockwise seasonal-trend-dispersion structure,
• WAVELET when you want a multi-scale signal-processing view.

Python example: SSA

import numpy as np
from detime import DecompositionConfig, decompose

t = np.arange(120, dtype=float)
series = 0.02 * t + np.sin(2.0 * np.pi * t / 12.0)

result = decompose(
    series,
    DecompositionConfig(
        method="SSA",
        params={"window": 24, "rank": 6, "primary_period": 12},
    ),
)

Observed output from examples/univariate_quickstart.py on the current docs build:

trend shape: (120,)
season shape: (120,)
residual shape: (120,)
backend: native

Published raw stdout:

• python_example_stdout.txt

Published method snapshot on the repo sample series

The current docs build also records a direct SSA versus STD snapshot on examples/data/example_series.csv.

Method	Backend	Trend std	Seasonal std	Residual RMS	Peak residual	Reconstruction error
SSA	native	0.6917	0.7036	0.0000	0.0000	0.0000
STD	native	0.6893	0.6558	0.0000	0.0000	0.0000

Published experiment record:

• method_snapshot.csv
• method_snapshot.json

This sample series is intentionally smooth and periodic, so both methods reconstruct it almost perfectly. Use the visual walkthroughs below when you want a noisier signal that leaves a visible residual.

Saving output with the CLI

python -m detime run \
  --method SSA \
  --series examples/data/example_series.csv \
  --col value \
  --param window=24 \
  --param rank=6 \
  --param primary_period=12 \
  --out_dir out/ssa_run \
  --output-mode summary \
  --plot

Published CLI stdout from the current docs build:

Running SSA on examples/data/example_series.csv...
Done. Results saved to out/ssa_run

Published output files:

• example_series_summary.json
• example_series_plot.png
• example_series_error.png
• command_stdout.txt

Published example outputs:

Where to go next

• Use Visual Univariate Walkthrough when you want one noisier signal with clearer residual structure.
• Use Visual Method Comparison when you want to compare SSA, STD, STDR, and STL on the same series before choosing a default.