You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A minor issue in the docs for the activations() function in src/layers/basic.jl.
If a beginner following the previous examples in the docs, e.g. loss(x, y) = crossentropy(m(x), y) + sum(norm, params(m))
and substituting activations() for params(), this will
result in the forward pass being computed twice, once when the model is evaluated, and again when activations is called.
The doc entry for activations() could provide an example or comment to emphasize that the model should not be called except through activations() itself.
The text was updated successfully, but these errors were encountered:
obviousnic
changed the title
activations() computes the forward pass from scratch
docs - add separate usage example for activations()
May 11, 2020
A minor issue in the docs for the activations() function in src/layers/basic.jl.
If a beginner following the previous examples in the docs, e.g.
loss(x, y) = crossentropy(m(x), y) + sum(norm, params(m))
and substituting activations() for params(), this will
result in the forward pass being computed twice, once when the model is evaluated, and again when activations is called.
The doc entry for activations() could provide an example or comment to emphasize that the model should not be called except through activations() itself.
The text was updated successfully, but these errors were encountered: