Changes between Version 4 and Version 5 of CodeCampIX/BreakOut1


Ignore:
Timestamp:
Mar 31, 2019 12:17:18 PM (6 months ago)
Author:
butler
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • CodeCampIX/BreakOut1

    v4 v5  
    7474 * Andrew has had student working on this. It is quite difficult. Key is the use of hypergeometric functions.  This might also be good for integration speeds in general (see below). 
    7575 * MLZ have been working on it too for a few simple models. 
    76   * They have some proof of concept in Born Again 
    77   * Need to work out API to work with SasView 
    78   * Plan for this ... 
    79    *Aim for next Code Camp to have BornAgain team integrate their code. 
    80   *Check !SasFit - Joachim may have implemented already. 
     76  * They have some proof of concept in !BornAgain 
     77  * Need to work out API to work with !SasView 
     78AGREED - Aim for next Code Camp to have !BornAgain team integrate their code. Also check !SasFit - Joachim may have implemented already. 
    8179 
    82 === 1D oriented data ... === 
    83  * How to fit a slice from a 2D data sets (oriented objects, magnetic scattering, etc.)  
     80=== 1D oriented data - How to fit a slice from a 2D data sets (oriented objects, magnetic scattering, etc.) === 
    8481 * Should we do it? 
    8582  * If we are going to provide live access to fast computers 
    8683  * If not too hard maybe do anyway .. a number of users with good cases 
    8784 * How would we do it? 
    88   * Make special 1D models? 
    89   * Or cut 2D models? 
    90  * Speed … ? 
    91   * Need to look at the speed ...= 
    92  * User interface … ? 
    93   * How to show the slicer options to the user? 
     85  * Make special 1D models 
     86  * Or use existing 2D models and just pull out appropriate cuts 
     87 * Using 2D would be better because 
     88  * Code exists and keeps us moving in the direction of full use of the data 
     89  * Can in fact properly match "smearing" from integrating over some angular range (e.g. 90 +/- 10 deg) 
     90  * BUT need to check if speed is fast enough this way 
     91 * How to present to the user (UI) 
     92  * Probably use existing slicer infrastructure 
     93   * How to show the slicer options to the user? 
     94   * Need to move slicer computational code out of GUI package (there is a ticket) 
    9495 * Narayan / Sylvain to find some example data. 
    9596  * Maybe start with simulation and treat as real data to test process 
    96  * AJJ to make some simulated data and share with everyone. 
     97AGREED - AJJ to make some simulated data and share with everyone. Sylvain to try scripting 1D from 2D to validate and check speed.  
    9798 
    9899=== Project to do benchmarking of optimisers === 
     
    102103   * Test !SasView optimisers against their defined problems 
    103104   * Identify classes of problems that would be representative for the SANS community 
    104   * AJJ - need to know the answer? 
     105  * AJJ - Do you need to know the answer? Else how to benchmark? 
    105106  * AJJ - how is SANS mathematically different from other test cases? 
    106107  * RKH - What about things where we have models that have multiple possible simulations? 
     108AGREED - Wojciech will try to get a better understanding of exactly what the !SasView community can help with. Simulated data would seem to be the best for this? 
    107109 
    108110=== Computation speed and integration options and approaches === 
    109 * Ingo : Suggestion - improve the integrations inside models. 
    110  * Get rid of nested integrals 
    111  * Use lebchev(?) Integrals? - some unrolled already- but other would require balancing competing requirements? 
    112  * How to include with GPU code ? 
    113  * Testing already … some models have numerical integration problems. Work underway. 
     111* Ingo : Suggestion - Can we improve the integrations inside models. 
     112 * Goal: to get rid of nested integrals 
     113  * Need to ensure that solution works with GPUs, multi cpu and headless modes of operatoin and is compatible with our BSD license so may be balancing competing requirements? 
     114  * Testing already … some models have numerical integration problems. Work underway. 
    114115 * Richard - will always have convergence issues for large particles. Or at least with particles that have one dimension that is massively larger than the other. 
    115 * Hypergeometric functions 
    116  * Can we implement them for more models? 
    117  * AJJ to find code from Johan. 
     116 * Use lebchev(?) Integrals? - some unrolled already- but other would require 
     117 * Hypergeometric functions (see Scatter discussion above) in principle can help with very fast integration but are they generic enough a solution to be implemented in models of !SasView? 
     118  
     119AGREED - AJJ to find code from Johan on hypergeometric integrals. Then get summer student to demonstrate use in some existing models along with performance enhancement metrics.