@fchollet Backprop the gradient to the feature? I've been thinking about plotting gradients to see which input pixels matter to my convnet.
What's the best / most economical way to estimate how relevant to a classification task a certain feature is?
-
-
-
@modeless@fchollet In which case you may likehttp://arxiv.org/abs/1312.6034
End of conversation
New conversation -
-
-
@fchollet permutation based importance in Random Forests is interesting (not sure how well it works outside of RF): https://www.stat.berkeley.edu/~breiman/RandomForests/cc_home.htm#varimp …Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet something similar to tf-idf from NLP? Each classification being a doc.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet If you use random forests, there are standard feature importance estimates http://alandgraf.blogspot.com/2012/07/random-forest-variable-importance.html?m=1 …Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet measure Information Gain?Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet train a model without it. See how much worse it isThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet if using svm/log reg, square the feature weights and rankThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet this is easy to do in simple linear regression where you can look at the coefficients and the p values.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@fchollet This is the paper for you: http://people.ee.duke.edu/~lcarin/brown12a.pdf …Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.