SampleSome

DSL-like data sampling library for Python 3.

Example - Zoo dataset

Let's assume that we need to generate a dataset containing of animals in the zoo. Each animal has: Id, ZooName, Type, Name, DaysInZoo, Weight, Age, Popularity (number of guests per day). Samplesome library provides a fluent-style API to generate a table of data. THere are three steps:

1. Create a model;
2. Configure fields by providing name, type and generator;
3. Generate data;

Create a model

Firstly, creating a model is as simple as:

animal = Model("Animal")

Configure fields

We can use the following field types:

• String
• Integer
• Float

Autoincrement field

Lets generate values like: A_1, A_2, ...

animal.Field("Id", String().Autoincrement("A_", 1))

Constant value generator

Zoo name will be fixed:

animal.Field("ZooName", String().Const("Jardin des plantes"))

Random choice from a set of values

animal.Field("Type", String().Choice(['Bear', 'Kangoo', 'Pig', 'Snake', 'Cat']))

Weights also can be provided.

Derivative fields

Derivative field computed from the fields declared before. We will generate names for animals comprising some random alias and animal type, which was already generated.

from random import choice
aliases = ['Angry', 'Shiny', 'Chatty', 'Sad']
animal.Field("Name", String().Derivative(lambda fields: f"{choice(aliases)} {fields['Type']}"))

Parametrized probability distibution

animal \
    .Field("Popularity", Integer().Exponential(scale=20.0)) \
    .Field("DaysInZoo", Integer().Normal(mu=150, sigma=50))

Conditional fields

Condition is a special type of derivative field, which allows defining a predicate and corresponding generator. We know that the weight of an animal depends on its type.

animal.Field("Weight", Float()
  .Condition(lambda fields: fields["Type"] == 'Bear', lambda f: f.Normal(mu=50, sigma=10, minf=10))
  .Condition(lambda fields: fields["Type"] == 'Kangoo', lambda f: f.Normal(mu=30, sigma=5, minf=5))
  .Condition(lambda fields: fields["Type"] in ['Pig', 'Cat'], lambda f: f.Normal(mu=5, sigma=1, minf=1))
  .Condition(lambda fields: fields["Type"] == 'Snake', lambda f: f.Exponential(scale=2, minf=1))
  .Condition(lambda fields: True, lambda f: f.Const(NAN)))

Take into account that min and max values may be limited to [minf:maxf].

Probability distibution can be defined in form of KDE

animal.Field("Age", Float().Kde([1.2, 2.3, 3.3, 3.2, .4, 5.6, 1.1, 1.0, 2.0, 2.1, 3.5, 6.7, 4.4, 7.7, 2.2, 5.5, 6.6, 1.2, 3.4, 9.0, 0.1, 3.2, 1.3, 2.5, 2.7], minf=0))

Data generation

Generate 1000 animals and export as csv:

animal.GenerateToCsv(1000, "zoo.csv")