4 ms·
What I find interesting is the implicit assumption that the underlying function being learned is differentiable or continuous to begin with. That's not always t
by FisDugthop 7y ago
What I find interesting is the implicit assumption that the underlying function being learned is differentiable or continuous to begin with. That's not always the case; for example, we often work with "categorically labeled" discrete binning problems.
- fxtentacle 7y agoIn those cases, people tend to use something like a categorical cross-entropy loss where you assign a continuous likelihood score to each discrete possibility, thereby making things differentiable again. https://www.tensorflow.org/api_docs/python/tf/nn/sparse_softmax_cross_entropy_with_logits https://www.tensorflow.org/api_docs/python/tf/nn/sparse_soft...
- julienreszka 7y agoData normalization is the 101 of any respectable machine learning course.
- jcims 7y agoYeah but it’s still enough to stall and hinder newbies (me) that deal primarily with categorical data until you can start to intuitively map the continuous back into the discrete. Nature of the beast it seems but still kind of a pain.