Cost Per View

Submitted by Kris Neuharth - 5/4/2018

Installation

This application requires Python 3, here I assume OSX for simplicity but this is easy to verify on other operating systems:

$ brew install python3
$ python --version

Next clone or unzip the project:

$ git clone [email protected]:krisneuharth/cost_per_view.git

Install dependencies and set path

$ pip install -r requirements.txt
$ python setup.py develop

Running the application

$ cd cost_per_view
$ python cpv.py

Output

{
  "cpv": {
    "rotation": {
      "Prime": 3.45,
      "Afternoon": 4.64,
      "Morning": 2.59
    },
    "rotation_by_creative_by_day": {
      "Morning - TEST001H - 2016-02-02": 3.5,
      "Afternoon - TEST002H - 2016-01-02": 5.29,
      "Prime - TEST001H - 2016-01-02": 3.64,
      "Prime - TEST002H - 2016-02-02": 2.33,
      "Morning - TEST001H - 2016-01-02": 1.72,
      "Afternoon - TEST001H - 2016-01-02": 3.64,
      "Prime - TEST002H - 2016-01-02": 5.29
    },
    "creative": {
      "TEST002H": 3.91,
      "TEST001H": 2.95
    },
    "rotation_by_day": {
      "Morning - 2016-01-02": 1.72,
      "Prime - 2016-01-02": 4.64,
      "Afternoon - 2016-01-02": 4.64,
      "Prime - 2016-02-02": 2.33,
      "Morning - 2016-02-02": 3.5
    },
    "day": {
      "2016-02-02": 2.8,
      "2016-01-02": 3.85
    }
  }
}

Running the tests

$ nosetests

Assumptions

0. I assumed Python would be ok for this exercise. I didn't use any of the data science tools but that may have made it easier.

1. I assumed that timezones didn't matter since it wasn't specified.

2. There are overlapping times for the rotations so I treated them as two separate spots for purposes of calculations.

3. I treated any unspecified rotation as "Other". This wasn't present in the data set but with a larger one it would show up.

4. I assumed a console print out would be sufficient for output.

5. I calculated a few other dimensions along the way just because.

6. I structured this app so that it would be fairly simple to make it run in parallel with multiprocessing.

7. I assumed that rounding to the nearest hundredth would be sufficient but this is easily changed.

8. I assumed a pretty clean data set and did not do much error handling.

9. I assumed larger data sets so I built this with memory use in mind, hence the classes with slots.

10. Possibly naively assumed that CPV is spend / views.

Thank you!
Kris