btlorch

icon company Universität Innsbruck icon location Innsbruck icon location Researcher with focus on multimedia security

Results 2 comments of btlorch

Multi-class classification with softmax likelihood

Thanks for the hints. The method `tf.nn.sparse_softmax_cross_entropy_with_logits` expects categorical labels, i.e., integer values in the range `[0, num_classes)`. If I interpret [the line](https://github.com/GPflow/GPflow/blob/8e0744af8717e0e341a92fc8fd509bc4e939c8fb/gpflow/quadrature/deprecated.py#L200) `D_out = Y.shape[1]` correctly, the quadrature method...

[Bug] Log transform not applied when computing posterior in heteroskedastic GP model

Thanks for looking into this issue. > [...] your suggested change seems reasonable, but you will want to take out the `Log` outcome transform in that case. But if the...