RigStatisticsCalculator::mobileVolumeWeightedMean() accumulated into a sum
from an out-parameter that callees were not required to write. Every
implementation of the per time step overload has early-return paths that
leave the value untouched, as does RigWeightedMeanCalc::weightedMeanOverCells,
so indeterminate stack memory could be summed and presented in the UI.
Return std::optional<double> from the computation layer instead of using an
out-parameter, making "value not available" explicit and impossible to ignore:
- RigWeightedMeanCalc::weightedMeanOverCells
- RigStatisticsCalculator::mobileVolumeWeightedMean and all overrides
Time steps without a computable mean no longer contribute to the accumulated
sum, and are excluded from the divisor. Previously they would either add
garbage or, with a naive zero initialisation, bias the average towards zero.
RigStatisticsDataCache keeps the existing HUGE_VAL marker on its public API
via value_or(), so behaviour seen by RigHistogramData and the overlay info
is unchanged apart from the fix.
Also:
- Set m_isVolumeWeightedMeanCalculated in the per time step cache, which was
never assigned, so the cached value was recomputed on every call.
- Remove a duplicated ensureKnownResultLoaded() call in
RigEclipseNativeVisibleCellsStatCalc.
Fixes#14766
The single-pass sum-of-squares formula subtracts two large and nearly equal numbers. Where a result such as PORO is identical or near-identical across the realizations, the true variance is below the rounding noise floor, the radicand turns negative and sqrt() returns NaN.
Compute the deviation in two passes instead, centering on the exact mean. This also removes the loss of significance the old formula had for values with a large common offset.
Min, max, range, mean and deviation all report HUGE_VAL when the input holds no valid values, but the sum reported the zero it was accumulated from. A caller testing the result with isValidNumber would accept that zero as real data.
Add BinningMode (Linear/Logarithmic) and OutOfRangeHandling (Exclude/Include in
Boundary Bins) to RigHistogramCalculator, with defaulted constructor arguments so
existing callers are unaffected. All four histogram plot data sources get shared
options for binning mode and an optional user-defined bin range through new PDM
fields on RimHistogramDataSource.
The grid statistics data source computes non-default histograms through a new
non-caching RigStatisticsDataCache::computeHistogram() pass-through, leaving the
cached statistics used by the 3D overlay and the P10/P90/mean annotation lines
untouched.
Selecting logarithmic binning enables logarithmic scale on the plot X axis once,
via a new signal handled by RimHistogramCurve.
Also fix the bin index computation to use uniform bins of width range/nBins,
consistent with calculatePercentil() and the bin edges drawn by the plots. The
previous divisor (nBins - 1) shifted counts by up to one bin width relative to
the drawn bars and slightly biased histogram-based percentile estimates.
When fewer than 9 valid realizations exist at a time step, the linear
interpolation percentile formula has no valid result for P10 and P90.
Previously HUGE_VAL (inf) was returned. Now clamp to min/max of the
available sorted values instead, avoiding inf in plotted curves.
* Refactor and extend percentile calculation utilities
Refactored percentile and quantile calculation logic for improved flexibility and maintainability. Added helper functions for quantile/percentile validation, introduced a new vector-based calculatePercentiles method, and updated calculateStatisticsCurves to use it. Refactored nearest-rank and interpolated percentile methods to accept percentiles as vectors and improved parameter validation. Enhanced documentation throughout and updated function signatures in the header for clarity.
Kristian Bendiksen
Magne Sjaastad
Gaute Lindkvist