17 | %hide Data.Container.Base.Morphism.Definition.DependentLenses.(=%>)
19 | ||| A loss is a parametric map whose parameter is the label
20 | ||| It is not a `Model` because there is no concept of initialisation
26 | ||| Run two losses in parallel, add their results
32 | ||| The loss of a coproduct of choices. When the types don't match, gradient
33 | ||| is infinite. In our examples we don't expect this to happen; but loss type
34 | ||| should be refined to exclude it eventually
46 | ||| The loss variant of `resolveByLabel`. Given a loss on resolved choices we
47 | ||| can produce a loss on an effectful output, where the ground-truth label
48 | ||| selects the effect
49 | ||| This means that the training loop does not need to handle any effects
50 | ||| anymore
72 | ||| The payoff object is the rank-0 tensor, not `Double`
87 | SoftargmaxCrossEntropyLogits