Outlier-search algorithm based on issue #385

[blazeiro]

Following the ADC irregularities for charge-injections found in Lund for some boards, I created an algorithm that scans parameter-time-scan data for outliers. The script assumes the irregularities to be less frequent than the "normal" results, and hence requires at least 10 samples per time-point for comparison. Using this data, the algorithm calculates the ADC mode for each time-point, as seen in the figure below.

Next, the script checks the ADC codes above and below the mode, one by one. If a gap of 2 or more is found (2+ ADC codes are not contained by any sample at the specific time-point), the points that follow are marked as an "outlier" or "potential outlier" cluster. Gaps of 1 are ignored to account for natural fluctuations or irregularities that can be found in any board, but this can be adjusted in the code. The resulting analysis can then be saved as a csv, with outlier frequency noted per sample. This can be used for quick board-evaluation, as the false-positive rate seems to be fairly low (about 1% from the tests I've done).
Taking the board from #385 as an example, the samples are split as follows.

The algorithm provides additional options for plotting and analysis. The samples can be viewed one-by-one with the "SINGULAR" option for plots:

A phase-analysis function (with plots) is also included:

Lastly, a somewhat deprecated method for finding outliers is left as an option. This method (the threshold scan), focuses on the ADC differences between subsequent time-points, marking differences larger than an arbitrary threshold as outliers, and does not require multiple samples. This method is fairly flawed, especially since it requires setting arbitrary thresholds, but I believe it could still be occasionally useful (though I can remove it).

[tomeichlersmith]

Thank you :) plots look good