search

adamwhat / koober

0 stars 0 forks source link

Adopt multinomial logistic regression #9

Closed Wang-Siyuan closed 7 years ago

Wang-Siyuan commented 7 years ago

Since we cannot impose strict 0 boundary in the default loss function(quadratic), we cannot actually guarantee the output to be always positive. This is an issue since we should never allow negative demand prediction.

One way to solve it is through multinomial logistic regression, which limits the possible output values. Related spark lib: http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.mllib.classification.LogisticRegressionWithLBFGS

Wang-Siyuan commented 7 years ago

Added a new branch for multinomial logistic regression. Resolving.