biogeme.expressions.weighted_logsum_exp module

Arithmetic expressions accepted by Biogeme: weighted log-sum-exp.

Michel Bierlaire

class biogeme.expressions.weighted_logsum_exp.WeightedLogSumExp(terms, weights)[source]

Bases: Expression

Weighted log-sum-exp expression.

It represents

\[\log\left(\sum_i w_i \exp(x_i)\right).\]

By convention, if the weighted sum is exactly zero, the expression returns zero, consistently with logzero(MultipleSum(...)).

Parameters:

  • terms (list[ExpressionOrNumeric])

  • weights (list[ExpressionOrNumeric])

deep_flat_copy()[source]

Deep flat copy of the expression.

Return type:

WeightedLogSumExp

get_value()[source]

Evaluate the value of the expression.

Return type:

float

recursive_construct_jax_function(numerically_safe)[source]

Generate a JAX function for weighted log-sum-exp.

Terms and weights are each compiled into one stacked JAX function. This keeps the weighted log-sum-exp node closer to a vector primitive and avoids maintaining one runtime list for all children together.

Return type:

Callable[[Array, Array, Array, Array], Array]

Parameters:

numerically_safe (bool)

recursive_construct_pymc_model_builder()[source]

PyMC builder for WeightedLogSumExp.

Return type:

PymcModelBuilderType

property terms

Terms \(x_i\).

property weights

Weights \(w_i\).

biogeme.expressions.weighted_logsum_exp.weighted_logsumexp(terms, weights)[source]

Factory function for weighted log-sum-exp.

Return type:

Expression

Parameters: