berkay-yalin / promethium Goto Github PK

Instead of raising exceptions within the "validate" functions when invalid arguments are used with the distribution functions, the nan value should be returned.

The reasoning behind this decision is to support large scale operations on arrays.

This makes identifying invalid arguments much more difficult, as the current "validate" functions provide detailed descriptions for why the arguments are rejected, so a more effective solution is welcome for discussion.

Example "validate" function, where each distribution function has its separate "validate" function:

Currently, the cumulative probability functions only have the upper bound as a parameter (k in both cases), and use 0 for the lower bound.

https://github.com/berkay-yalin/python-probabilities/blob/c9c9e55a444bec83d71221a26fa5fe181d8c669f/python_probabilities/distributions_binomial.py#L21-L22

https://github.com/berkay-yalin/python-probabilities/blob/c9c9e55a444bec83d71221a26fa5fe181d8c669f/python_probabilities/distributions_poisson.py#L21-L22

This approach works well when calculating probabilities below a certain number (e.g., P(X≤7)), but becomes more complicated when trying to calculate a range of values (e.g., P(4≤X≤7)).

In this case, to calculate it using the currently available functions we would have to calculate the following P(X≤7) - P(X≤3).
For example, if we were using a binomial distribution of X ~ B(11, 0.33), we would calculate P(4≤X≤7) like this:

from python_probabilities import *
print(Bcd(7,11,0.33) - Bcd(3,11,0.33))

Which gives us an output of 0.5096391069522413707200.

Therefore, I suggest adding an optional lower parameter to the cumulative probability functions, which would default to 0 if not passed into the function. This would make it easier to calculate the probability of a range of values without the need to subtract two probabilities. Additionally, this would increase the efficiency of the algorithm, as it would avoid the need for the functions to iterate twice through the same probabilities.

Inconsistent return types

Some of the distribution functions return a float whilst others return a decimal.Decimal.
All distribution functions should be consistent in their return types.

Inconsistent parameter naming

Inconsistent naming for parameters of various functions where the actual parameters represent the same thing.
For example, InvBinomialCD uses y whereas InvGeometricCD uses area:

https://github.com/berkay-yalin/python-probabilities/blob/d8585c2c41af3152a365e387ccb5b974b265aeca/python_probabilities/distributions_binomial.py#L59-L61

https://github.com/berkay-yalin/python-probabilities/blob/d8585c2c41af3152a365e387ccb5b974b265aeca/python_probabilities/distributions_geometric.py#L30-L35