matlab - Combine output score to form a better score -

matlab - Combine output score to form a better score -

i have 10 scheme perdition outputs only, yes outputs.

all outputs scored taken rmse value against ground-truth value:

rmse = sqrt( sum( (true_value(:)-predictions(:)).^2) / numel(true_value))

one rmse 1 system, eg: 0.355 scheme s has 1000+ predictions.

from side of view, without knowing ground-truth values, how create utilize 10 outputs , score improve prediction.

if have 10 systems outputs y_i , rmse value of rmse_i, index i goes 1 10. linearly weighted output y corresponds to:

in matlab, weights each signal can obtained with:

w = (1-rmse) ./ sum(1-rmse);

for example, if have rmse values:

rmse = [0.355, 0.1672, 0.7479, 0.2756, 0.0046, 0.4870, 0.1925, 0.9554, 0.8374, 0.1426];

the weights each signal is:

w = [0.1105, 0.1427, 0.0432, 0.1242, 0.1706, 0.0879, 0.1384, 0.0076, 0.0279, 0.1469];

it can seen outputs higher rmse value have lower weight, i.e. info less reliable.

matlab prediction