Round learner map coordinates for better summarizing in review.

[GoogleCodeExporter]

What steps will reproduce the problem?
1. View content statistics for the oppia.org Exploration "Welcome to Oppia" for 
the state named "Finding Helsinki".

What is the expected output? What do you see instead?

Expect: summary of commonly selected locations.
Actual: the coordinates shown are so precise they can't be summarized. 
(examples below)

What operating system are you using?
OSX / Chrome

Please provide any additional information below.
- See attached screenshot. The coordinates provided by Google Maps API have so 
many decimal places of precision that they can't be summarized to a useful "top 
5" locations summary.  As such, none of the top answers match more than 1-2 
times despite 5,000+ learners reaching the page.  Example coordinate precision: 
45.460130637921004,46.845703125

Suggested solutions:
1) Easier: perhaps consider rounding coordinates to the nearest hundreds or 
thousands places (though would that be losing too much precision for cases 
desiring street addresses?)
2) Better: rather than displaying coordinates (which may not have much 
immediate meaning to human reviewers), if Google Maps API has the ability to 
return a "nearest known city" string it may be best... (though specific to this 
state which is focused on cities). Another option might be returning a "nearest 
known address" string if Maps API can do it. 

Original issue reported on code.google.com by anu...@google.com on 10 Jun 2015 at 6:26

Attachments:

• [Screen Shot 2015-06-10 at 2.22.18 PM.png](https://storage.googleapis.com/google-code-attachments/oppia/issue-816/comment-0/Screen Shot 2015-06-10 at 2.22.18 PM.png)

[GoogleCodeExporter]

That does get tricky when you consider the use case of clicking on street 
addresses...

The other angle would be to enhance the admin view somehow, such as:
1) Enable filtering the answer tallies by precision.
2) (Probably technical complicated) Display dots on a map, so the admin can 
visualize concentrations of where users clicked.

Original comment by peters...@google.com on 10 Jun 2015 at 6:56

[GoogleCodeExporter]

+Marcel

Thanks for raising this, Michael!

For information, Marcel is currently working on a framework for summarizing and 
visualizing answer data at the moment. The idea is to allow different 
summaries/visualizations for each interaction type, rather than the current 
one-size-fits-all "string diff then order by frequency".

Marcel: this might be a good use case to test out your framework. We can have 
the Map interaction show summary data in the form of a heatmap or based on 
answers rounded to fewer decimal places.

Some more random brainstorming:

  * all this probably depends on what sort of feedback it'd be useful to give the learner. Some ideas include:
    - wrong continent / wrong country
    - distance from target location (or several wrong locations). Could be interesting e.g. for training people how to follow directions in a map ("turn left at the crossroads", etc.).
    - characteristics of the location (is it a park, is it water)
  * we probably need one visualization for the first couple of cases (involving distances) and a second for the location characteristics
  * thinking about how the learner would select their answer, it actually seems reasonable to use the default zoom level of the map as a guide to figuring out what resolution is needed in the visualization. The learner is unlikely to choose locations far outside this window, so this actually gives us a good guideline of what can be considered a meaningful distance.

Marcel, do we currently include the interaction's customization args in the 
parameters we send to the calculation job? If the reasoning in the last bullet 
point above is good, this looks like a case where they would be useful.

Original comment by s...@google.com on 10 Jun 2015 at 7:22

[GoogleCodeExporter]

Quick note that I have to give props to Justin for raising this.
I merely documented his suggestion. Thanks again, Justin!

Original comment by anu...@google.com on 10 Jun 2015 at 7:31

[GoogleCodeExporter]

Hi,
Yes using the zoom level for this seems like a nice solution and I think this 
would be a great use case for us. 

We are storing params in answer_dicts, see  
https://code.google.com/p/oppia/source/browse/core/domain/event_services.py?name
=widget-answer-views-rebased#90 and 
https://code.google.com/p/oppia/source/browse/core/domain/stats_services.py?name
=widget-answer-views-rebased#217, so the calculations have access to them. Do 
you know if these params would contain interaction customization args like zoom 
level for map interactions?

Original comment by mschmittfull@gmail.com on 11 Jun 2015 at 1:46

[GoogleCodeExporter]

No, params are different. I think we have to store the customization args here, 
similar to how we persist the interaction id:

    https://code.google.com/p/oppia/source/browse/core/domain/stats_domain.py?spec=svndd6995e56c5372995d75bbf0a8f2af31d969431c&name=widget-answer-views-rebased&r=dd6995e56c5372995d75bbf0a8f2af31d969431c#94

    https://code.google.com/p/oppia/source/browse/core/domain/stats_domain.py?spec=svndd6995e56c5372995d75bbf0a8f2af31d969431c&name=widget-answer-views-rebased&r=dd6995e56c5372995d75bbf0a8f2af31d969431c#94

Can you do that?

Original comment by s...@seanlip.org on 11 Jun 2015 at 3:06

[GoogleCodeExporter]

Oh right -- yes I'll try to include that.

Original comment by mschmittfull@gmail.com on 11 Jun 2015 at 3:43

[GoogleCodeExporter]

Hi Marcel, just had one other thought about the proposed solution: having the 
customization args also allows us to show multiple choice answers that weren't 
clicked on (i.e. have 0 submissions)!

Original comment by s...@google.com on 13 Jun 2015 at 12:29

[GoogleCodeExporter]

Original comment by s...@google.com on 13 Jun 2015 at 12:29