Is it possible to trace the optimization history?

[GiggleLiu]

Is it possible to know the loss/parameters in each step (rather than each function call).

[Gnimuc]

I'm not sure, but if setulb doesn't provide such info in dsave, we may need to hack into the Fortran code.

[Gnimuc]

Just to make sure you didn't miss this example:

LBFGSB.jl/test/driver3.jl

Lines 151 to 154 in 5148b3c

	# "3) the value of the objective function f,"
	# "4) the norm of the projected gradient, dsave(13)"
	println("Iterate ", isave[30], " nfg = ", isave[34], " f = ", f, " |proj g| = ", dsave[13])

It looks like LBFGSB only exposes its APIs at the iteration level.

[GiggleLiu]

Thanks for the example, it really helps.