-
Notifications
You must be signed in to change notification settings - Fork 3
Laundry List of Losses and Penalties #1
Comments
An example for a unsupervised loss is the density level detection loss (DLD) However, I think even then practically speaking the types would be different to implement, no? All in all I am not sure about this as I haven't spend a lot of time dwelling on unsupervised losses. I simply wanted to keep the option open |
In our other comically long conversation (JuliaML/META#8 (comment)), we came up with the following types:
Cost
is a functionf(X, y, c)
wherec
can be anythingLoss
is a functionf(X, y, g(X))
, soc
is some function of the dataSupervisedLoss
doesn't needX
, so its a functionf(y, g(X))
UnsupervisedLoss
doesn't needy
, so its a functionf(X, g(X))
Penalty
doesn't needX
ory
, so its a functionf(W)
whereW
are some model params/weightsDiscountedFutureRewards
... @tbreloff can you edit this andIMPORTANT: everybody should edit this comment to check off
Costs
as they are done and also to add newCosts
This post a work-in-progress...
Supervised Losses: (implemented in Losses.jl)
L1DistLoss
L2DistLoss
LPDistLoss
LogitDistLoss
,L(y, t) = -ln(4 * exp(y - t) / (1 + exp(y - t))²)
L(y, t) = log(1+exp(-t*y))
Unsupervised Losses:
Penalties: (implemented in ObjectiveFunctions perhaps?)
Constrants: These are actually penalties. The penalty is "infinitely bad" when the constraint is violated.
lower[i] < x[i] < upper[i]
for all i)Combinations of Penalties: I would like an easy/automatic way of combining instead of defining by hand, but this may not be feasible.
Discounted Future Rewards:
@tbreloff -- halp!
Some sources:
The text was updated successfully, but these errors were encountered: