I read a little bit and decided to go for an approach using Median Absolute Deviation, and some factor around it. I prime the database with the known good ranges supplied in the assignment. Uploading data re-computes the acceptable ranges for the next upload. It seems the method I chose doesn't align that well with the data. A factor that's specific to each country data set might improve on it.
The database is currently recreated on each run of the server. I decided to use SQLlite for simplicity. The MAD calculation could and should be offloaded to the database.
The currency data is hard-coded. It should be fairly simple to fetch and store from an external source in an separate task.
No particular validation is done on the uploaded data. Little error handling is done on nonexistent data for a country. I'd ideally use some sort of framework for the first part.
I like to work with ORMs, so I tried out SQLAlchemy for the first time. It's a bit more involved than I'm used to. It strikes me as a poor choice for working with JSON data.
I included a set of stubs for a REST API and an example test case. I'll expand on these if I find the time.
Optionally install and activate a virtualenv:
virtualenv some_path
. some_path/bin/activate
pip install --upgrade pip # Necessary on my Ubuntu at least.
Install the dependencies:
pip install -r requirements.txt
In a shell, run:
python run.py
Open your browser and point it to:
http://127.0.0.1:5000/
You should now see an empty list of registered charges, and an initial set of distribution information. Choosing a file and clicking "Upload", should import charges from the file.
Use the filters to drill down on country and validity.
So far, only one test case is implemented as an example. Nose should find and execute this and any others added later:
nosetsts