ntpstats.series

Time-series container used throughout ntpstats.

Sign convention

offset always follows the ntpd / RFC 5905 convention::

offset = server_time - local_time      (seconds)

i.e. a positive offset means the local clock is behind the reference. Some sources use the opposite sign (chrony tracking/statistics logs, linuxptp); the parsers normalise on import so every series in this package can be compared directly.

TimeSeries dataclass

An offset time series plus optional auxiliary columns.

Attributes:

Name	Type	Description
t	ndarray	Sample times, POSIX seconds (float64, UTC).
offset	ndarray	Clock offset in seconds (server - local).
name	str	Human readable label (file name, peer address...).
source_format	str	Parser that produced the series (loopstats, chrony-tracking...).
extra	Dict[str, ndarray]	Additional per-sample columns (frequency_ppm, jitter, delay...), same length as t.
meta	Dict[str, object]	Free-form metadata (peer address, notes, warnings).

to_pandas()

DataFrame with a UTC DatetimeIndex (needs pandas; see :mod:ntpstats.adapters).

from_pandas(df, offset='offset', time=None, name=None, negate=False) classmethod

From a DataFrame (see :func:ntpstats.adapters.from_pandas).

to_parquet(path, compression='zstd')

Write as Parquet (needs pyarrow); every command reads it back.

select(mask)

Return a new series keeping only the samples where mask is true (mask may also be an integer index array).

sorted()

Return the series sorted by time with exact duplicate times removed.

between(start=None, end=None)

Restrict to start <= t <= end (POSIX seconds, either may be None).

median_interval()

Median sampling interval in seconds (the natural tau0).

regularity()

Fraction of intervals within 10 % of the median interval (1.0 = perfectly regular).

to_uniform(tau0=None, max_gap=3.0)

Resample onto a uniform grid for stability analysis.

Returns (grid_t, x, tau0) where x is the offset (phase) linearly interpolated on the grid. Grid points that fall inside a gap longer than max_gap * tau0 are set to NaN so that the stability estimators simply skip the terms that touch them instead of inventing data.

mjd_to_unix(mjd, seconds)

Convert (Modified Julian Day, seconds past midnight UTC) to POSIX seconds.