How to define a variable conditional on a state of a markov chain

[Former user]

I want to define a variable with different distributions based on the current state of a Markov chain.

For example, I define a weather variable using Markov chain as

weather = markov.chain_from_matrix(('sunny','rainy'), ('sunny',(0.95, 0.05)), ('rainy',(0.8, 0.2)))

I want to have a different distribution of temperatures according to today's weather

Potentially, I want something like

temp_dist = weather.switch({'sunny': p_sunny, 'rainy': p_rainy})

But Lea will return an error

AttributeError: 'Chain' object has no attribute 'switch'

[Pierre Denis]

A Markov chain is not a probability distribution, so switch method cannot be called. To get a probability distribution, you need to define an initial distribution like “yesterday was sunny, with x% chance” You can obtain easily a uniform initial distribution using .state attribute (see doc here)

>>> yesterday_weather = weather.state
>>> yesterday_weather
sunny : 0.5
rainy : 0.5

From this initial state, the Markov chain can build a distribution for today:

>>> today_weather = weather.state.next_state(1)

(or next days, replacing 1 by 2, 3, etc.) Then, you may define temperature for today using a switch (i.e. conditional probability table). For instance:

>>> p_sunny = 18 + Lea.binom(4,0.5)
>>> p_rainy = 10 + Lea.binom(4,0.5)
>>> today_temperature = today_weather.switch({'sunny': p_sunny, 'rainy': p_rainy})

As a sanity check,

>>> today_temperature.given(today_weather=='sunny')
18 : 0.0625
19 : 0.25
20 : 0.375
21 : 0.25
22 : 0.0625 

… that is the p_sunny temperature distribution, as expected.

[Pierre Denis]

hopefully clarified (not an enhancement/bug)!