perda.utils.diff#

perda.utils.diff.aggregate_timestamps(ts_list)[source]#

Concatenate timestamp arrays and count occurrences per unique timestamp.

Parameters:: ts_list (list[npt.NDArray[np.int64]]) – Each element is a 1-D int64 array of timestamps (one per event).
Returns:: (unique_timestamps, counts), both sorted by timestamp.
Return type:: tuple[npt.NDArray[np.int64], npt.NDArray[np.int64]]

perda.utils.diff.diff(rpi_data, server_data, timestamp_tolerance_ms=2, diff_rtol=0.001, diff_atol=0.001)[source]#

Compare two SingleRunData objects and report differences.

Performs a three-stage comparison:

Variable-name alignment: reports C++ names present in one run but not the other
Point-level diff: for each common variable, classifies every data point as a base-only extra, incoming-only extra, value mismatch, or match.
Summary + plot: prints a diff summary table and displays an interactive Plotly bar chart of per-bucket diff counts.

Parameters:

rpi_data (SingleRunData) – The reference (baseline) run.
server_data (SingleRunData) – The incoming run to compare against the baseline.
timestamp_tolerance_ms (int) – Maximum timestamp delta (ms) to consider two points as matching. Defaults to 2.
diff_rtol (float) – Relative tolerance for value comparison via np.isclose. Defaults to 1e-3.
diff_atol (float) – Absolute tolerance for value comparison via np.isclose. Defaults to 1e-3.

Return type:

Figure