mojito-mock 0.3.0

Mocking temporal data made easy !

Mojito

Mocking temporal data made easy !

There are lot of data mocking framework in Python. But none of them are really oriented toward generating statistically homogeneous data, especially temporal event. Mojito is designed for that !

How to use it

Start by installing Mojito pip install git+https://github.com/PPACI/mojito.git.

from datetime import datetime
from pprint import pprint

from mojito import DateGenerator, EventComposer, FixedValueGenerator, NumberGenerator, PropertyEventGenerator, \
    RandomChoiceGenerator

event1 = PropertyEventGenerator(
        properties={
            "timestamp": DateGenerator(center=datetime(2018, 5, 2, 0, 0, 0), deviation=3 * 3600),
            "age": NumberGenerator(mean=15, deviation=3, return_int=True),
            "gender": RandomChoiceGenerator(['M', 'F']),
            "label": FixedValueGenerator(1)
            })
event2 = PropertyEventGenerator(
        properties={
            "timestamp": DateGenerator(center=datetime(2018, 5, 10, 0, 0, 0), deviation=3 * 3600, skew=0),
            "age": NumberGenerator(mean=30, deviation=3, return_int=True),
            "gender": RandomChoiceGenerator(['M', 'F'], weights=[2, 1]),
            "label": FixedValueGenerator(0)
            })

composer = EventComposer()
composer.add_generator(event1, 3)  # 3 samples from this generator
composer.add_generator(event2, 2)  # 2 samples from this generator

pprint(composer.generate(shuffle=True))

Will output

[{'age': 16.0,
  'gender': 'M',
  'label': 1,
  'timestamp': datetime.datetime(2018, 5, 1, 22, 23, 1, 450298)},
 {'age': 19.0,
  'gender': 'F',
  'label': 1,
  'timestamp': datetime.datetime(2018, 5, 1, 21, 0, 11, 583775)},
 {'age': 30.0,
  'gender': 'M',
  'label': 0,
  'timestamp': datetime.datetime(2018, 5, 9, 22, 57, 30, 441924)},
 {'age': 15.0,
  'gender': 'F',
  'label': 1,
  'timestamp': datetime.datetime(2018, 5, 2, 5, 59, 54, 96498)},
 {'age': 32.0,
  'gender': 'M',
  'label': 0,
  'timestamp': datetime.datetime(2018, 5, 10, 0, 15, 55, 676862)}]

API

Mojito use a model where a PropertyEventGenerator will be used to generate sample events. An Event is something happening, characterized by the statistical distribution of the sample it represent. A Sample is the generated data. It could represent a visit on your site, or someone buying a specific item or whatever. Remember that if you want to generate sample from two statistical distribution, you will have to create two events and compose them as in the example. Currently, the main distribution used to generate data is the Normal distribution as it's used to represent lot's of real life distribution.

EventGenerator

A PropertyEventGenetator will be instantiated with a dictionary of {"name":PropertyGenerator}. This event generator will be at the center of your mocking task as it's describing how event should look like.

PropertyGenerator

You have access to the following PropertyGenerator

• DateGenerator
  • Output datetime distributed around the supplied center datetime
  • Distribution is a Skew Normal
  • You can pass skew=0 to have a non skewed normal distribution
  • scale is in second, so deviation=3600 will result in a standard deviation of 1 hours around the provided datetime
• FixedValueGenerator
  • Always output the same value
• RandomChoiceGenerator
  • Take random choice from a provided list of possibilities
  • You can pass weights=[a, b] to weight the list accordingly
• NumberGenerator
  • Output number distributed around the supplied mean
  • Distribution is a Skew Normal
  • You can pass skew=0 to have a non skewed normal distribution
  • You can pass return_int=True to generate integer instead of float
  • Also available as NormalNumberGenerator class
• PoissonNumberGenerator
  • Output number from a Poisson distribution
  • Only distribution parameter is mu, the average number of event in the time period
  • You can pass return_int=True to generate integer instead of float

Composition

Real models are aggregation of multiple, different, events. To simulate this, you can use the EventComposer.

composer = EventComposer()
composer.add_generator(event1, 3)  # 3 samples from this generator
composer.add_generator(event2, 2)  # 2 samples from this generator

Add your EventGenerator and the number of wanted generated events from each generator. You can also remove one with .remove_generator(event1).

• .generate() will return you a list of generated events as dictionary
• .to_csv("output.csv") will save the generated events as csv, ready for your process !