ErrorStats

Summary of one error signal over its finite samples, in the signal's own SI unit.

Full docstring

`mean`, `median`, `min`, `max` and `final` keep the error's sign; `rms`, `max_abs` and
`p95_abs` are of its magnitude. WHY this set: evo's rmse / mean / median / std / min / max
(the SLAM-evaluation standard), plus what tracking needs on top: the worst magnitude, a p95
that ignores a single glitch sample, and `final`, the error left when the motion ends.
Missing data is NaN with ``n == 0``, never 0.

`iae`, `ise` and `itae` are the control-loop integral criteria (∫|e|, ∫e², ∫(t - t₀)|e|, in
the signal's unit × s, unit² × s and unit × s²): trapezoids over the intervals whose both ends
were observed. WHY NaN by default: a series given without its time axis has no integral, and
data serialised before these fields existed still loads.

Import

from robot_msgs.compare import ErrorStats

ROS 2no ROS equivalent

Fields

Name	Type	Default
`n`	`int`	`required`
`rms`	`float`	`required`
`mean`	`float`	`required`
`median`	`float`	`required`
`std`	`float`	`required`
`min`	`float`	`required`
`max`	`float`	`required`
`max_abs`	`float`	`required`
`p95_abs`	`float`	`required`
`final`	`float`	`required`
`iae`	`float`	`nan`
`ise`	`float`	`nan`
`itae`	`float`	`nan`

Example

One instance from the package's own test catalog; the output is its real repr().

ErrorStats(n=3, rms=0.0021602468994692866, mean=0.0013333333333333333, median=0.002, std=0.0016996731711975948, min=-0.001, max=0.003, max_abs=0.003, p95_abs=0.0029, final=0.003, iae=nan, ise=nan, itae=nan)

to_dict() as JSON

{
  "_type": "compare/ErrorStats",
  "n": 3,
  "rms": 0.0021602468994692866,
  "mean": 0.0013333333333333333,
  "median": 0.002,
  "std": 0.0016996731711975948,
  "min": -0.001,
  "max": 0.003,
  "max_abs": 0.003,
  "p95_abs": 0.0029,
  "final": 0.003,
  "iae": NaN,
  "ise": NaN,
  "itae": NaN
}