I’m using Optim.jl to train a model but it keeps overfiting. is there nice way of implementing a early stopping callback that has the current x? I tried setting the callback in Optim.Options but i only get the iteration, loss and time. is it possible to access other optimizer states?

Early stopping with Optim when the validation loss increases

pkofod January 7, 2021, 10:17pm 2

Yes, it should work if you set extended_trace=true in the options

1 Like

Topic		Replies	Views
Optim.jl callback Optimization (Mathematical)	1	1656	January 26, 2021
Optim callback function -- how to use optimization variable Optimization (Mathematical) optim	1	1295	January 26, 2021
Optim L-BFGS() - Can I save internal state? Optimization (Mathematical) question , optim	1	1764	January 25, 2018
Plotting the evolution of the optimization using Optim.jl General Usage plotting , optim , forwarddiff	3	936	June 25, 2020
[ANN] EarlyStopping.jl 0.1.0: For terminating iterative algorithms Statistics	0	461	February 21, 2021