Transcript Emma report
Last Two Weeks • Parallel Environment – SUR BLUEGENE • Account set up • DVC can’t cross-compile since Bluegene computing node doesn’t support dynamic linking while currently DVC adopt a dynamic plugin loading method – Department Opteron Cluster • • • • Group project space granted Get 64-bit PATH Solver for DVC Dakota successfully configured and compiled Parallel running of Dakota tested Last Two Weeks • Calibration using Dakota – Update simulation code to calculate calibration error on-the-fly – Set up Dakota pattern search optimization case (a Local Optimization Method) to do the calibration – A few minutes to get a local optimal set of parameters for a single video – A group of video : see the next page Last Two Weeks • Calibration using Dakota for a group of videos – Wrote a dvc Parallel Wrapper to parallel run dvc on different experiment data Dvc Wrappers Processor 0 Dispatch Master Sum up Errors Processor 1 Processor 2 Processor 3 Processor 4 DVC DVC DVC DVC video1 video2 video3 video4 continued – Took about one and a half hour to get a result for four videos, hopefully this time won’t increase too much if more videos are added – My first result matches the data about the same well as previous result from Matlab, though different parameter value Next • Tune up the parameters of the local optimization method and try more initial values • Try Global optimization method in Dakota and probably mix both : Global optimize as a start and use the result as initial of local • Continue the work of parameter sensitivity study