Does it matter if I set the two classes to be 1 and 0, instead of 1 and -1? For some reason, the accuracy is way higher when I implement concise Softmax regression with the two classes being 1 and 0.
Related Topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
Hw3 [1.4] | 3 | 473 | February 11, 2019 | |
L1 regularization implementation in Gluon
|
0 | 391 | March 24, 2020 | |
Mod.score inconsistency between LogisticRegressionOutput and SoftmaxOutput | 8 | 1119 | May 3, 2018 | |
Softmax Regression in Gluon | 11 | 1830 | December 27, 2019 | |
Environment | 7 | 788 | January 12, 2020 |