Weight Types

Each of these algorithms is parameterized over several different semirings. The weight_type argument is used to specify the semiring.

Name	\(\oplus\)	\(\otimes\)	\(\bar{0}\)	\(\bar{1}\)	\(\mathbb{S}\)/dtype
LogViterbi	\(\max\)	\(+\)	\(-\infty\)	0	float32
Viterbi	\(\max\)	\(*\)	0	1	float32
Real	\(+\)	\(*\)	0	1	float32
Log	logsum	\(+\)	\(-\infty\)	0	float32
Boolean	or	and	false	true	uint8
Counting	\(+\)	\(*\)	0	1	int32
MinMax	\(\min\)	\(\max\)	\(-\infty\)	\(\infty\)	float32

Invariants

Check the semiring properties.

import pydecode.test.utils as test_utils
graph, weights, weight_type = test_utils.random_setup()

Check the additive and multiplicative identities.

assert (weight_type.Value.one() * weight_type.Value(weights[0])).value == weights[0]
assert (weight_type.Value.zero() + weight_type.Value(weights[0])).value == weights[0]