planrich / tu_aic_13 Goto Github PK

It seems no proper logging facilities are utilized to output informational or debug messages. Instead we use quite a few print statements throughout.

Python offers a standard logging module, which we should use. Of interest: http://victorlin.me/posts/2012/08/good-logging-practice-in-python/

this is an important task for the 28th of november.
it needs to be accurate, interesting presentation that
is split into 3 parts:

• an introduction what we have achieved
• what our implementation does
• demo of our tool

write a small selenium tests project, where we can simulate a lot of wokers solving tasks

Both applications (main, crowd) should be deployable in a simple, mostly automated manner, preferably into a stable, productive state. Not only for the sake of making development easier but also for the TA to be able to easily get the application running locally.

This means optionally pre-populating the database with known-good data, using Apache and mod_wsgi instead of the built-in application server, not relying on Cron, and logging properly. It also means the applications should be deployable outside and independently of OpenShift and work just as well (or better?).

Issue #17 is related to this, as are issues #23 and #24.

I'd start with making the applications deployable using Vagrant, so that anyone interested can just do a vagrant up and a few minutes later have a working deployment of our main and crowd applications running in two small virtual machines.

load the statistics from the local database and display them

currently the data is just random

The tutor should be able to setup our system easily

is there a presentation yet? If not it would be a good time to create one.
Also i think we should focus on presenting the tool well. ( @KonradMSteiner already started to do cleanup and add data )

do we meet before the presentation? I think that would not be a bad idea, but only if there is already a presentation + concept to present tool.

Is there someone to coordinate this?

or at least delete the keyword

use \b in python
http://docs.python.org/2/library/re.html

we should limit the text, so that we don't see the whole text (which is quite long sometimes)

we should provide a Button in the list_tasks site, where we can solve the task directly

Reason: now we have the problem, that we always get random numbers and we can't finish the rating, because of the amount of data we get all the time.

so a simple Button, called "Solve" with a link to the ID will be enough.
Based on Richards approach on marking already solved tasks in the session, we should not display these tasks in the list_tasks site (I don't know if we already implemented it)

"Start with the updates to your crowdsourcing model. Firstly, think about how you can measure the quality of your human workers. Once you have a metric for this, blocking notoriously bad workers is trivial over the MobileWorks API"

define the metrics! implement it!

don't run it every minute, an hour or so is enough

f.e. http://main-tuaic13.rhcloud.com/query/Apple?d=1
apple should also be possible

"Afterwards design your dynamic pricing solution. Think about what your general approach is going to be, and implement it."

This is an optional feature. I would still recommend to implement it.

highlight searched keyword in the solve_task page
maybe in javascript

• how often was it mentioned
• ...

use taps or so to display the graphs (not all of them on one page)

like google analytics

fill the keywords in the dbs on startup

"Add a “garbage collection” mechanism that tracks which tasks are never being picked up, and removes them."

Define what is the resonable decision here and implement it.

change the crowd source application that then clicking 'solve a random task' it really
takes a random task not the first.

The open_tasks list on crowd app crashes when you want to show all open tasks(more than 100000) at once. We should only show a limited number of open_tasks in one page and allow scrolling through this pages.

list the open tasks with paginagtion

It would be nice to keep track of the date from when we are analyzing some keyword and also from when a worker is in our system. I suggest we add a datetime column to the worker and keywords tables.

Instead of Cron the applications should use Advanced Python Scheduler to run periodic jobs like the scraper or garbage collector. This offers greater control over running job instances, makes deployment easier (no external components) and allows for better cross-platform compatibility.

one cannot solve a task on http://crowd-tuaic13.rhcloud.com/solve_task?r=t

Internal error 4! Could not finish task. Here is a new one!

As far as I remember this is due to the fact that the main application does not take the answer!

"Finally, build a simple Web interface for your whole service. The Web interface should be both, a management interface for inspecting the sentiment data and the quality of your workers, as well as a GUI for issuing queries to your service and visualize the results (e.g., via diagrams or plots). You may use your choice of platform for building this Web interface."

do research on the crowd sourcing topic

produce a paper