[SOLVED] Statistics on AdvancedTimer

[Sarah]

Hi!

While not directly related. I thought my question about the statistics retrieved from the Advanced timers may be useful here.

My questions concerns the interpretation of the data.
My question cencerns what the various data types represent? What does NUM entail? Is there any reason that the TOTAL of the TOTAL is a factor of 100 larger then the mean, are they multiplied by NUM, which corresponds most of the time? And are the units of the TOTAL then in ms or those of the mean?

When collecting the data you get the following, as retrieved from the Sofa documentation:

Steps Duration Statistics (in ms) :
LEVEL   START    NUM      MIN     MAX   MEAN     DEV    TOTAL  PERCENT ID
  0       0     100      86.57  127.50  109.99    7.77 10999.1  100    TOTAL
  1       0.06    1      86.32  127.25  109.75    7.75  109.75   99.78 .Mechanical
  2       0.09    1      79.42  112.21   99.97    7.02   99.97   90.89 ..Build linear equation
  3       0.09    1       0.84    1.36    1.14    0.14    1.14    1.04 ...forces in the right-hand term
  3       1.34    1       0.07    0.14    0.10    0.02    0.10    0.09 ...shift and project independent states
  3       1.44    1       0.39    0.68    0.55    0.08    0.55    0.50 ...local M
  3       2.23    2       7.52   17.67   12.72    2.34   25.44   23.13 ...J products
  3      27.70    1       0.28    0.54    0.41    0.06    0.41    0.37 ...local K
  3      28.13    1      54.07   79.54   70.61    5.53   70.61   64.20 ...JMJt, JKJt, JCJt
  3      98.75    1       0.88    2.29    1.31    0.24    1.32    1.20 ...implicit equation: scaling and sum of matrices, update right-hand term 
  2     100.06    1       2.71    4.79    3.75    0.51    3.75    3.41 ..Solve linear equation
  1     109.84    1       0.01    0.02    0.02    0       0.02    0.01 .UpdateMapping
  2     109.84    1       0       0       0       0       0       0    ..UpdateMappingEndEvent
  1     109.85    1       0.09    0.28    0.14    0.03    0.14    0.12 .UpdateBBox

I was trying to find more infromation on these outputs online but was having a hard time finding answers to these possibly rather trivial questions. Help would be appreciated!

Kind regards,

Sarah

[Sarah]

Hi!
Was just wondering if it would be better to open my previous post as a new topic. Would be great to get to understand the Advanced Timer more.

Best, Sarah

[Hugo]

Hi @sarah

I moved your question in a new topic.
To be honest, I am not super familiar with the details of these stats.

I would say that:
– 100 is the number of steps which are run and the digits in the table are actually average over these 100 iterations
– NUM means how many times this is computed
– TOTAL which is multiplied by 100 (10999.1) I would say this is an error, and should be 109.991)

It would be nice if you would be curious and check into the code, and even put some fix there! I am as knowledgeable as you are here 🙂

Still I hope this helps a bit!
Cheers

Hugo

[Sarah]

Hi Hugo,

I looked into some of the code of the AdvancedTimer.cpp. This is what I found:

json TimerData::createJSONArray(int s,json jsonObject, StepData& data)
{
    double value = 0;
    ctime_t ttotal = stepData[AdvancedTimer::IdStep()].ttotal;

    // Level :
    value = strToDouble(getVal(data.level), 4);
    jsonObject["Level"] = value;

    // Start
    value = strToDouble(getTime(data.tstart, data.numIt), 4);
    jsonObject["Start"] = value;

    // Num
    value = strToDouble(getVal(data.num, (s == 0) ? 1 : nbIter), 4);
    jsonObject["Num"] = value;

    // TMin
    value = strToDouble(getTime(data.tmin), 4);
    jsonObject["Min"] = value;

    // TMax
    value = strToDouble(getTime(data.tmax), 4);
    jsonObject["Max"] = value;

    // Mean
    double mean = (double)data.ttotal / data.num;
    value = strToDouble(getTime((ctime_t)mean), 4);
    jsonObject["Mean"] = value;

    // Dev
    value = strToDouble(getTime((ctime_t)(sqrt((double)data.ttotal2/data.num - mean*mean))), 4);
    jsonObject["Dev"] = value;

    // Total
    value = strToDouble(getTime(data.ttotal, (s == 0) ? 1 : nbIter), 4);
    jsonObject["Total"] = value;

    // Percent
    value = strToDouble(getVal(100.0*data.ttotal / (double) ttotal), 4);
    jsonObject["Percent"] = value;

    return jsonObject;
}

So the Total (ttotal) is right at least, that the mean is the one that takes the num into consideration. It would seem most clear if num represented the number of itteration but it is also remarkable that ‘nbIter’ is a value and that in the folowing part of the code also introduces ‘numIt’. So am a bit confused about if these three values (num, nbIter, numIt) are referring to the same thing? Also I don’t quite understand yet why NUM is different for the different components in the example, as unless it just did multiple itterations of one of the component they should be the same right. And this makes me wonder wheter it is actually referring to the itterations of the simulation as a whole, or to something different.

    class ValData
    {
    public:
        int num, numIt;
        double vmin;
        double vmax;
        double vtotal;
        double vtotal2;
        double vtotalIt;
        int lastIt;
        ValData() : num(0), numIt(0), vmin(0), vmax(0), vtotal(0), vtotal2(0), vtotalIt(0), lastIt(-1) {}
    };

So far I ran my simulation, printing the log at every itteration and with that all my NUM are equal to 1, whilst with that avoiding the problem I don’t think the question has completely been answered yet. The interpretation of the statistics may be a useful question or comment to add in the documentation, in particular for the example placed there, as I think this could help many. Thank you! Hope that this is of use to people in the future.

Best,
Sarah

[Hugo]

Hi @sarah

It’s good to dare looking in the code.
Do never hesitate to propose also improvement to the doc yourself, the documentation is open for contribution as well.

I will try to work on it, since the topic is fresh.

Hugo

[epernod]

Hi @sarah,

the output table is indeed not very user friendly. In fact the 1st line should be read separately from the rest of the table.
The first line is showing mean values over 100 iterations of your simulation except for the “Total” which is the total in ms of the 100 iterations.

Then if you look at the rest of the table, this is the detail (of the 1st line) for one simulation step. So there the NUM is the number of time the operation is done per simulation step. You see that “Mechanical” and “UpdateBBox” are the 2 mains operations of your step and if you sum their % you nearly get the 100%, here I think the visual/draw operation is not shown.

Also, I don’t know what is your usage of the advancedTimer but there is a graphical version of it:
See full PR here.
I apologize that this is not mentioned in the doc… it is on my todo list.

Erik

[Hugo]

Hey

@epernod I added some text and image here about our discussion and about your profiler.
Do not hesitate to complete it.

Shall I close the topic for now?
Best

Hugo

[Sarah]

Hi Hugo and Erik!
Thanks for all the tips, has been helpful. Also it is very cool to see the documentation and other options expand. Yes, I’ll close the topic. I think it has been resolved.

Best,

Sarah

[Author]

Posts