Investigate drop in Seq Learning performance

GateNLP / gateplugin-LearningFramework

A plugin for the GATE language technology framework for training and using machine learning models. Currently supports Mallet (MaxEnt, NaiveBayes, CRF and others), LibSVM, Scikit-Learn, Weka, and DNNs through Pytorch and Keras.

https://gatenlp.github.io/gateplugin-LearningFramework/

GNU Lesser General Public License v2.1

26 stars 6 forks source link

Investigate drop in Seq Learning performance #26

Closed johann-petrak closed 8 years ago

johann-petrak commented 8 years ago

The unit test result in TestEngineMalletSeq has dropped from 0.3646/0.4299 to 0.2965/0.3472 after we worked on the feature extraction code. Find out what is going on there.

johann-petrak commented 8 years ago

This was mainly because of how we did not correctly use boolean features and also did not correctly use the correct missing value treatment in the older version, so the newer unit test values were actually the correct ones, even though worse.