samirelanduk / inferi Goto Github PK

inferi is a statistics and data science Python library.

>>> import inferi
>>> variable = inferi.Variable(11, 45, 23, 12, 10)
>>> variable.mean
20.2
>>> variable.variance()
219.7

inferi can be installed using pip:

$ pip3 install inferi

inferi is written for Python 3, and does not support Python 2. It currently supports Python 3.6.

If you get permission errors, try using sudo:

$ sudo pip3 install inferi

The repository for inferi, containing the most recent iteration, can be found here. To clone the inferi repository directly from there, use:

$ git clone git://github.com/samirelanduk/inferi.git

inferi currently has no dependencies, compiled or otherwise.

inferi is a tool for performing basic statistical analysis on datasets. It is pure-Python, and has no compiled dependencies.

The fundamental unit of inferi data analysis is the Variable. It represents a set of measurements, such as heights, or favourite colours. It is not the same as a Python variable - it represents variables in the statistics sense of the word.

>>> import inferi
>>> heights = inferi.Variable(178, 156, 181, 175, 178)
>>> heights
'<Variable (178, 156, 181, 175, 178)>'
>>> heights.length
5

If you like, you can give the variable a name as an appropriate label:

>>> heights = inferi.Variable(178, 156, 181, 175, 178, name="heights")
>>> heights.name
'heights'

You can also give an existing sequence, such as a list, and the result will be the same:

>>> heights = inferi.Variable([178, 156, 181, 175, 178])
>>> heights
'<Variable (178, 156, 181, 175, 178)>'

Values can be accessed by indexing:

>>> weights.values
(12, 19, 11)
>>> weights[0]
12
>>> weights[-1]
11
>>> weights.max
19
>>> weights.min
11

Variables have the basic measures of centrality - mean, median and range.

>>> heights = inferi.Variable(178, 156, 181, 175, 178)
>>> heights.mean
173.6
>>> heights.median
178
>>> heights.mode(
178

See the full documentation for details on Variable.mean, Variable.median, and Variable.mode. Note that if the variable has more than one mode, None will be returned.

Variables can also calculate various measures of dispersion, the simplest being the range:

>>> heights = inferi.Variable(178, 156, 181, 175, 178)
>>> heights.range
25

You can also calculate the variance and the standard deviation - measures of how far individual measurements tend to be from the mean:

>>> heights.variance()
101.3
>>> heights.st_dev()
10.064790112068906

By default the Variables will be treated as samples rather than populations, which has consequences on the value of both the variance and the standard deviation. To get the population values for each, simply set this when you call the method:

>>> heights.variance(population=True)
81.04
>>> heights.st_dev(population=True)
9.00222194794152

Again, see the full documentation of Variable.range, Variable.variance, and Variable.st_dev for more details.

It is often useful to compare how two variables are related - whether there is a correlation between them or if they are independent.

A simple way of doing this is to find the covariance between them, using the Variable.covariance_with method:

>>> variable1 = inferi.Variable(2.1, 2.5, 4.0, 3.6)
>>> variable2 = inferi.Variable(8, 12, 14, 10)
>>> variable1.covariance_with(variable2)
0.8033333333333333

The sign of this value tells you the relationship - if it is positive they are positively correlated, negative and they are negatively correlated, and the closer to zero it is, the more independent the variable are.

However the actual value of the covariance doesn't tell you much because it depends on the magnitude of the values in the variable. The correlation metric however, is normalised to be between -1 and 1, so it is easier to quantify how related the two variable are. Variable.correlation_with is used to calculate this:

>>> variable1 = inferi.Variable(2.1, 2.5, 4.0, 3.6)
>>> variable2 = inferi.Variable(8, 12, 14, 10)
>>> variable1.correlation_with(variable2)
0.662573882203029

Usually, more than one thing is measured in an experiment, and so you would have more than one variable. For example, you might ask someone's name, their age, their height, and whether or not they smoke. Each of these four metrics is a variable:

>>> variable1 = inferi.Variable("Jon", "Sue", "Bob", name="Names")
>>> variable2 = inferi.Variable(19, 34, 38, name="Ages")
>>> variable3 = inferi.Variable(1.87, 1.67, 1.73, name="Heights")
>>> variable4 = inferi.Variable(False, True, True, name="Smokes")

These can be combined into a single Dataset as follows:

>>> dataset = inferi.Dataset(variable1, variable2, variable3, variable4)
>>> dataset.variables
(<Variable 'Names' ('Jon', 'Sue', 'Bob')>, <Variable 'Ages' (19, 34, 38)>, <Va
riable 'Heights' (1.87, 1.67, 1.73)>, <Variable 'Smokes' (False, True, True)>)

A dataset can be thought of as representing a table of data, where each variable is a column. This dataset represents a table like this:

Names Ages Heights Smokes

Jon   19   1.87    No
Sue   34   1.67    Yes
Bob   38   1.73    Yes

You can get the rows of a dataset too:

>>> dataset.rows
(('Jon', 19, 1.87, False), ('Sue', 34, 1.67, True), ('Bob', 38, 1.73, True))

A Dataset can be sorted, by default by the first column but this can be made otherwise:

>>> dataset.sort()
>>> datset.rows
(('Bob', 38, 1.73, True), ('Jon', 19, 1.87, False), ('Sue', 34, 1.67, True))
>>> dataset.sort(variable3)
>>> dataset.rows
(('Sue', 34, 1.67, True), ('Bob', 38, 1.73, True), ('Jon', 19, 1.87, False))

Probabilty is a way of looking all the ways something can happen and assessing how likely the outcomes are.

Everyone's favourite example is rolling a die - there are six possible outcomes in the Sample Space:

>>> space = inferi.SampleSpace(1, 2, 3, 4, 5, 6)

This defines a sample space with six outcomes. Each of these is a simple event:

>>> space.simple_events
{<SimpleEvent: 1>, <SimpleEvent: 2>, <SimpleEvent: 3>, <SimpleEvent: 4>, <Simp
leEvent: 5>, <SimpleEvent: 6>}
>>> space.event(5)
<SimpleEvent: 5>
>>> space.event(5).probability()
0.16666666666666666
>>> space.event(5).probability(fraction=True)
Fraction(1, 6)
>>> space.chances_of(5)
0.16666666666666666

Events are some combination of simple events. For example, to define the event that a rolled die produces an even number:

>>> even_event = space.event(lambda o: o % 2 == 0, name="even")
>>> even_event
<Event: even>
>>> even_event.name
'even'
>>> even_event.probability()
0.5
>>> even_event.outcomes()
{2, 4, 6}
>>> even_event.outcomes(p=True)
{2: 0.16666666666666666, 4: 0.16666666666666666, 6: 0.16666666666666666}
>>> even_event in space
True

Two events can be compared. Here we create two more events:

>>> odd_event = space.event(lambda o: o % 2 != 0, name="odd")
>>> large_event = space.event(lambda o: o > 4)
>>> odd_event.mutually_exclusive_with(even_event)
True
>>> large_event.mutually_exclusive_with(even_event)
False
# Does knowing number is even affect chances of being odd? (Obviously...)
>>> odd_event.dependent_on(even_event)
True
# Does knowing number is even affect chances of being greater than 4?
>>> large_event.dependent_on(even_event)
False

You can even make new events from them...

>>> small_and_even = large_event.complement & even_event
>>> small_and_even.probability()
0.333333333333333
>>> small_and_even.outcomes()
{2, 4}

1 May 2018

• Implemented combinatorics and permutations.
• Added basic probability tools:
  • Events, simple events and event spaces.
  • Conditional probability.
  • Concept of 'and' and 'or'.
• Turned certain property methods into actual properties.

6 October 2017

• Added Dataset class for collating Variables.

27 August 2017

• Renamed Series 'Variable'
• Added error handling.
• Added Variable averaging and adding/subtracting.
• Added z-score.
• Generally overhauled everything.

26 March 2017

• Added option to make a Series a population rather than a sample.
• Added covariance and correlation measures.

21 March 2017

• Added basic Series class.
• Added methods for measures of centrality and basic measures of dispersion.