ProcessWindowEvents¤

Descriptive statistics of a signal per time window, and the windows whose mean or spread changed against the previous window or stands out from all windows.

ts_shape.events.engineering.process_window · Engineering guide · API reference

Example¤

import ts_shape
from ts_shape.events.engineering.process_window import ProcessWindowEvents

df = ts_shape.make_timeseries(["mixer_torque"], n_points=1440, freq="1min", baseline=40.0, noise=0.5)
df.loc[700:760, "value_double"] += 3.0

pw = ProcessWindowEvents(df, "mixer_torque")
stats = pw.windowed_statistics(window="1h")
print(pw.detect_mean_shift(window="1h", sensitivity=2.0))
variance = pw.detect_variance_change(window="1h", ratio_threshold=2.0)
outliers = pw.window_comparison(window="1h")

Methods¤

All arguments are keyword-only.

Method	Returns
`windowed_statistics(window='1h')`	per window: `count`, `mean`, `std`, `min`, `max`, `median`, `p25`, `p75`, `range`
`detect_mean_shift(window='1h', sensitivity=2.0)`	`prev_mean`, `current_mean`, `shift_sigma`
`detect_variance_change(window='1h', ratio_threshold=2.0)`	`prev_std`, `current_std`, `variance_ratio`
`window_comparison(window='1h')`	per window: `mean`, `z_score_vs_global`, `is_anomalous`

Notes¤

A mean shift is flagged when |current mean − previous mean| > sensitivity × the previous window's standard deviation.
A variance change is flagged when current std / previous std is above ratio_threshold or below its reciprocal.
In both, start is the previous window's start and end the current window's start.
is_anomalous: the window mean lies more than 2 standard deviations of all window means from their average.