Summary
Summary: a task in LegoFlow-SWE (Harbor dataset). The outlier scores computed from features with the OutOfDistribution class are not properly normalized to the 0,1 range. As a result, the distribution of scores depends unpredictably on the chosen n neighbors parameter, making it hard to interpret…
The task
The outlier scores computed from features with the `OutOfDistribution` class are not properly normalized to the [0,1] range. As a result, the distribution of scores depends unpredictably on the chosen `n_neighbors` parameter, making it hard to interpret scores or set thresholds.
Part of Lego-X/LegoFlow-SWE.