Circle or Square or Circle within a square (Feature #27)

red sea wrote:

The form of boxes, shading to answer..

Proposed addition to allow the user to choose between:
-Circle
-Square
-Circle within a square

best regards

It appears to me that a circle or oval shape is filled in much easier by the students. So I think this could help.

thanks
i got this error messege:

LaTeX Error: Environment tikzpicture undefined.
Undefined control sequence.
Illegal unit of measure (pt inserted).
Missing number, treated as zero.
Illegal unit of measure (pt inserted).
Illegal unit of measure (pt inserted).
Missing number, treated as zero.
Illegal unit of measure (pt inserted).
Undefined control sequence.

after i added :
\usepackage{tikz}

to the head, it worked with (shape=oval) only.
when i try to used (shape=circle or squere) i got this message :

Package xkeyval Error: value `circl' is not allowed.

regards

shape must be square or oval. You can get a circle using shape=oval with the same value for both width and height.

ok thanks , now worked.

can i use "diamond" or "Star" or any shape in tikz?

best regards

can i use "diamond" or "Star" or any shape in tikz?

No. Drawing the shape is not enough to make it work. Some work has to be done for counting the number of black pixels inside the shape. This work has only be done for squares and ovals.

thanks.

regards

I just tested (I had not noticed you already had an implementation).

1. The oval shape is quite nice and will help the candidates to fill out the marks much easier.
2. The pdf rendering with pdflatex is much slower. With a 30 question rendered for two students (using csvsimple) using the oval command takes 7.3 seconds real time, with the default rectangular boxes 0.601 seconds, both timed with time pdflatex source.tex. This would make it very
   slow indeed. My normal exams have between 40 and 80 candidates.

Observing the log output produced while pdflatex makes its run makes me guess that most of the time is used rendering the ovals. I appears to pause at the end of each answer sheet. It has to put 131 ovals per answer sheet.

Would it not be much quicker if you rendered the oval once and re-use it every next time? Both LaTeX and pdf can do define and reuse.

Of course I will have to test the scanning performance and quality and will do that asap.

In any case, it look promising.

Would it not be much quicker if you rendered the oval once and re-use it every next time? Both LaTeX and pdf can do define and reuse.

I'm afraid I don't see how to implement this idea in the automultiplechoice.sty package... Please send a patch if you know how to do this.

Took a look into your implementation and considered the render and reuse.

It might be possible using the \newsavebox, \savebox and \usebox commmand but then you would have to draw 'box' and content differently.
I would consider such to much of a change in the inner workings of the latex sty file. That is not the way to go I think.

I also made and experiment and replaced the 4 tikz arc commands and cycle with one ellipse.
In the exam set I did the experiment on, (16 open questions, 2 open questions with in total 71 tik boxes per form) rendering an exam for 59 students,
it reduced the compilation time (one pdflatex run) from 2 minute 24 seconds to 1 minute 8 seconds.

Of course some tweeking might be needed to get the ellipse in the proper proportion, but since it is somehow the closed figure we want, that might do the trick.

At the moment consider it low priority and leave the code as is.

To give you the complete results of my experiment, here you have the ellipse in latex/tikz code, with the arc parts commented out.
I made a copy of the .sty file and edited that before changing anything in your sources.

Would like to upload my oval definition to share it with you.
\renewcommand the \AMC@shape@oval command in a separate .sty file, to experiment with it first.

Would like to upload my oval definition to share it with you.

Sure. Please attach it (click on the "files" banner at the bottom of the page when adding notes).

I just did two experiments with the oval/ellipse form type.

1. Filling in the form by the students is both faster and cleaner (Filling a round shape with a pen/pencil can be done much quicker)
2. Scanning and scoring works quite well. I tried hard to fill the shapes properly and have an MSE of 1.6 and a sensitivity of 5.7.

So from my point of view the oval is a success and can be used (in my next exam series).
I gladly accept the slower production speed (it only costs my patience, a nice exercise anyway) in exchange for more comfort and reliability at the filling out the form side.

For the time being, I will stick to your original oval, which works best in scanning and the GUI (r 1197)

Very nice: this also cut by two the processing time with ovals on my computer. I will replace the current implementation with yours shortly in the development code.

Even nicer would be to have the option to use the default rectangular boxes in the questions and the oval boxes only on the separate answer form, where they have play out their benefits when filling in by pencil or pen.
This could again save a substantial off the processing time, because to all questions the choice boxes are rendered twice, once with the question and once on the separate form.

Something like \AMCOvalOnAnserformOnly or similar more appropriate to the current API style.
Or \AMCQuestionBoxShape and \AMCAnswerformBoxShape, both defaulting to rectangular.

Maybe you might also default to rectngualr boxes in the solution rendering. No one has to fill in the solution anyway ;-))
This will again save some time in the 'update documents' process.

Maybe you might also default to rectangular boxes in the solution rendering. No one has to fill in the solution anyway ;-))
This will again save some time in the 'update documents' process.

I just tested (I had not noticed you already had an implementation).

1. The oval shape is quite nice and will help the candidates to fill out the marks much easier.
2. The pdf rendering with pdflatex is much slower. With a 30 question rendered for two students (using csvsimple) using the oval command takes 7.3 seconds real time, with the default rectangular boxes 0.601 seconds, both timed with
   

   time pdflatex source.tex
   

   
   This would make it very slow indeed. My normal exams have between 40 and 80 candidates.

Observing the log output produced while pdflatex makes its run makes me guess that most of the time is used rendering the ovals. I appears to pause at the end of each answer sheet. It has to put 131 ovals per answer sheet.

Would it not be much quicker if you rendered the oval once and re-use it every next time? Both LaTeX and pdf can do define and reuse.

Of course I will have to test the scanning performance and quality and will do that asap.

In any case, it look promising.

Would it not be much quicker if you rendered the oval once and re-use it every next time? Both LaTeX and pdf can do define and reuse.

I'm afraid I don't know how to do that, particularly if letters or other characters are to be drawn inside the shapes. Please write the LaTeX code if you can.

Alexis Bienvenüe wrote:

Thanks.
I used your idea to write a patch that should work in all contexts: can you test it? Load this package right after automultiplechoice.

I will try it tomorrow at work on the same machine I did the timing on. The machine I am now at is too fast ;-)).
Even on this fast machine, running latex on the current exam, two passes for both exam and solution takes a while and is of course only able to use one core per run.

Come to think of it: On a halfway modern machine, having more then one core (this one has six, my faithful laptop, the one I timed with two), you could run the corrections and exam latex run in parallel (ssuming the latex run is error free), which would chop the default 'update documents' process in halve, because both runs would write to different files (I think). Starting two tasks in parallel in the shell is easy enough, as you probably know.

Pieter Van den Hombergh wrote:

Alexis Bienvenüe wrote:

Thanks.
I used your idea to write a patch that should work in all contexts: can you test it? Load this package right after automultiplechoice.

I will try it tomorrow at work on the same machine I did the timing on. The machine I am now at is too fast ;-)).
Even on this fast machine, running latex on the current exam, two passes for both exam and solution takes a while and is of course only able to use one core per run.

Come to think of it: On a halfway modern machine, having more then one core (this one has six, my faithful laptop, the one I timed with two), you could run the corrections and exam latex run in parallel (ssuming the latex run is error free), which would chop the default 'update documents' process in halve, because both runs would write to different files (I think). Starting two tasks in parallel in the shell is easy enough, as you probably know.

I could not resist: It feels a lot quicker.
On this machine (at home, 6 core amd processor with way to much memory so disk buffers are never filled), same exam-conditions as the time measurement above the result is

time pdflatex source.tex 
real    0m18.967s
user    0m18.605s
sys    0m0.276s

I also ran the 'update documents' process in the GUI again ( on my home machine), and now it is also quick again (with its 4 passes), so this is probably the way to go for this feature. It would cause me to bring down my coffee ration.

Both renderings (Question and Solution) look optically fine, did not yet test the scanning and post processing.

Good work and very quick turnaround on this one. Bravo.

By the way, this morning the students score the exam I used in the timing example. That rendering scanned flawless and have my first result this way.
There where almost no incompletely filled boxes, because these ovals are a lot easier to fill with a pencil (preferred) or felt-tip pen.

It works.¶

As promised my test results:

First run, old style ovals timing.
real    3m0.869s
user    2m59.883s
sys    0m0.404s

Second run, savebox ovals timing
real    0m25.377s
user    0m24.738s
sys    0m0.388s

which is more then 6 times better!

I also tested with a complete cycle: prepare, layout detection, print, scan and score and this is also Ok (I scored one exam by hand with a pencil and scanned that. Should give maximum grade of course).

I had the small issue that the corrected version did not have filled boxes, which I did not notice at home, so I must have a further look into that. I will let you know if this problem is on my side.

Conclusion: Works.¶

• On both home workstation and laptop your first and second patch both seem to produce the same result.
• Is there anything that I should look into or pay attention to in particular?
• My observations of different behaviour between laptop and computer at home was wrong. The experiments had a different set up so it was probably not something in the first patch
• Anyway, This looks good.
• Will use it for my next exams coming week.

Thanks, as always.

I do not want to open another issue, but was wondering if a complete latex run is required when you select Update marking scale in the marking tab.
I just assume that you want/need to re-acquire the questions weight (points) and correct/wrongs, but need not do all the regeneration for all candidates.
We do pre-association, and if you need only information en not the student information, then it might be less time consuming if you processed the material
for one candidate only, thereby gathering al information about the questions, which will be the same for all candidates.

I added it here because it relates to the speed of latex processing in some way. Not doing unneeded work is always faster, whatever algorithm you choose ;-))
One could also imagine that such run would be done in some kind of DRAFT mode, e.g. avoiding complex renderings which might save some processing time.

Maybe my assumption is wrong and do you need to recompute all xy values.

I noticed that the program makes two or three (or maybe four) latex runs when using Update marking scale, always using only one cpu. I could imagine that you need two runs for each processing variant, but it might be the two 'exam' runs and the two 'answer' runs could be run in parallel, as long as they use different output (and aux) files, but you could simply put those in separate subdirectories using pdflatex options such as -output-directory=DIR and -jobname name. This could also produce some speedup, at least when you have more then one core to spend.

Alexis Bienvenüe wrote:

was wondering if a complete latex run is required when you select Update marking scale in the marking tab.

I'm afraid yes. All candidates may have different questions, different answers, different scoring strategies (the case of random numerical values with a \AMCnumericChoices is a good example). This is not always the case but we have to take these cases into account.

imagine that such run would be done in some kind of DRAFT mode

Yes, this is a good idea... but I don't know how to do that.

I noticed that the program makes two or three (or maybe four) latex runs when using Update marking scale

This looks strange. It should be 2 runs at most. Can you confirm that ?

I will be preparing exams for next next week, starting preparations today, so I will have a look and record my findings here.

I rand an exam today and tested a bit with 'update marking' while also looking at the time stamps of the files.

I am running 1.1.1335 (svn:1346) (svn update, make deb and install, ubuntu 12.04).

From what I see, in Marking-> check 'update marking scale', the program only updates the amc-compiled.* files and no other. I think this is correct because as you explained a complete latex run is in order.
So either my earlier observation is wrong or some other files are no longer updated, in particular the DOC-catalog.pdf (nice addition by the way) and the DOC-corrige.pdf are not updated in 'updating marking scale'.

Watching the computer activity during a long run (30 questions, 40+ candidates, update marking) I saw only one pdflatex process as in one run, which means that this is also quite optimal.
I am running the program on the same machine and it appears a lot quicker. Your work must have helped, because my machine is at the same OS level as before.

Good. Thanks for the report!