The emission probability matrix is defined using an MxC matrix:
(
)
Like the transition probability, the sum of the emission probabilities is 1. That is,
∑
Various attributes/properties of Hidden Markov Model have been identified in the above comments.
Prediction of
subsequent states is the ultimate goal for any statistical model/algorithm. However, before making predictions, we need to
solve
two
main problems in HMM.
Assessment problem;
A learning problem.
Decoding problem.
Assessment problem
First, we define the model (θ) as follows:
θ→s,v,a_ik,b_jk
Considering
the model
and the sequence of observed signs (
)
we
need to determine
probability of the
generation of a certain sequence of states / signs generated (determined) based on the
model.
There can be many {θ
1
,θ
2
…θ
n
} models.
We need to find such
𝑝
and correctly
classify the sequence
using
Bayes rule
. In this case, the following equality is relevant:
|
|
Learning problem
In general, HMM is an unsupervised machine learning process where different types of visible characters are known.
But the number of hidden cases is unknown. The idea behind the HMM is to try different
options and require more
computation and processing time. Therefore, HMM uses study data and a certain number of
hidden states to make faster,
better predictions.
After determining the high-level structure of the model (the number of hidden and visible states), it is necessary to
estimate transition (a_ij) and emission (b_jk) probabilities using training sequences. This is noted as the learning
Dostları ilə paylaş: