Multinomial

Multinomial distribution for categorical data.

Parameters

• events

  Type → dict | list | None

  Default → None

  An optional list of events that already occurred.

• seed

  Default → None

  Random number generator seed for reproducibility.

Attributes

• mode

  The most likely value in the distribution.

• n_samples

  The number of observed samples.

Examples

from river import proba

p = proba.Multinomial(['green'] * 3)
p = p.update('red')

p('red')

0.25

p = p.update('red').update('red')
p('green')

0.5

p = p.revert('red').revert('red')
p('red')

0.25

You can wrap this with a utils.Rolling to measure a distribution over a window:

from river import utils

X = ['red', 'green', 'green', 'blue', 'blue']

dist = utils.Rolling(
    proba.Multinomial(),
    window_size=3
)

for x in X:
    dist = dist.update(x)
    print(dist)
    print()

P(red) = 1.000
<BLANKLINE>
P(red) = 0.500
P(green) = 0.500
<BLANKLINE>
P(green) = 0.667
P(red) = 0.333
<BLANKLINE>
P(green) = 0.667
P(blue) = 0.333
P(red) = 0.000
<BLANKLINE>
P(blue) = 0.667
P(green) = 0.333
P(red) = 0.000
<BLANKLINE>

You can wrap this with a utils.Rolling to measure a distribution over a window of time:

import datetime as dt

X = ['red', 'green', 'green', 'blue']
days = [1, 2, 3, 4]

dist = utils.TimeRolling(
    proba.Multinomial(),
    period=dt.timedelta(days=2)
)

for x, day in zip(X, days):
    dist = dist.update(x, t=dt.datetime(2019, 1, day))
    print(dist)
    print()

P(red) = 1.000
<BLANKLINE>
P(red) = 0.500
P(green) = 0.500
<BLANKLINE>
P(green) = 1.000
P(red) = 0.000
<BLANKLINE>
P(green) = 0.500
P(blue) = 0.500
P(red) = 0.000
<BLANKLINE>

Methods

call

Probability mass/density function.

Parameters

• x — 'typing.Any'

revert

Reverts the parameters of the distribution for a given observation.

Parameters

• x — 'typing.Hashable'

sample

Sample a random value from the distribution.

update

Updates the parameters of the distribution given a new observation.

Parameters

• x — 'typing.Hashable'