How to set performance weights for crossentropy in patternnet?

4 次查看(过去 30 天)
I am training a patternnet on a two-class imbalanced training set, about 20/80.
The crossentropy function (when called directly) allows for specific performance weights to be assigned, and I want to use that to overweight penalties for the rare class.
However, I don't see how I can set performance weights when using crossentropy in a patternnet. I can assign net.performFcn to be 'crossentropy', and have two properties under net.performParam, 'regularization' and 'normalization'.
So, how can performance weights be specified?

回答(1 个)

Greg Heath
Greg Heath 2018-11-13
See both
help crossentropy
and
doc crossentropy
In the latter see the section on
perfWeights - performance weights
Hope this helps
*Thank you for formaly accepting my answer*
Greg

类别

Help CenterFile Exchange 中查找有关 Sequence and Numeric Feature Data Workflows 的更多信息

产品


版本

R2018b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by