Opened 16 months ago

Closed 3 months ago

#297 closed defect (migrated)

Sudden jump uphill after taking small steps

Reported by: timholy Owned by: ipopt-team
Priority: normal Component: Ipopt
Version: 3.12 Severity: normal
Keywords: Cc:

Description

I have an optimization problem for which Ipopt takes a sudden jump uphill, without fanfare, at iteration 15. I've posted a log from a print_level=12 run here: https://gist.githubusercontent.com/timholy/48f6b631a67c430ba2cd653d98b2163b/raw/8b56a38493250c4c9978662644b89fd1a53c93f3/optimization.log. (I'll try to attach this file, too.) You can see I ran with the derivative checker and several other options. The end result is that the solution after iteration 15 (or however many you want to allow it) is almost 10x worse in terms of objective function than the initial starting point.

It would be a bit challenging to share the code for this problem (it's written in Julia, based on a large codebase), but briefly it's an image registration problem. The parameters of the optimization are just a rotation angle and 2 translations. There's a quadratic interpolation performed on the image pixels, which means that the objective function is C1 but not C2. The edge pixels might even violate the C1, technically speaking. However, you'll note the derivative-checker doesn't complain.

Attachments (1)

register.log (923.8 KB) - added by timholy 16 months ago.

Download all attachments as: .zip

Change History (2)

Changed 16 months ago by timholy

comment:1 Changed 3 months ago by stefan

  • Resolution set to migrated
  • Status changed from new to closed

This ticket has been migrated to GitHub and will be followed up there: https://github.com/coin-or/Ipopt/issues/297

Note: See TracTickets for help on using tickets.