Attribute selection


there is a way to select (automatically with some sort of algo or function in deeplearning4j or other libs) the best attributes from a set of attributes?

the example like this one

it’s clear that col1 it’s useless, it’s always 0.

but other scenario can also apply, where an attribute “influence” it’s not important for the classification. It’s possible to determinate a set of best attributes to use and make the network work better?

@ramarro123 that’s a bit outside the scope of the framework. You typically use something like chi-squred tests or PCA for that kind of thing. You may want to look at something like smile for that:

It isn’t automatic, but if you are mostly looking to understand your data better in order to decide which features to drop, you can use the analysis feature in data vec.

For example see: Quickstart with Deeplearning4J – dubs·tech