mirror of
https://github.com/janishutz/eth-summaries.git
synced 2026-09-10 19:15:25 +02:00
[IML] Examples
This commit is contained in:
@@ -83,6 +83,20 @@ $$
|
||||
\end{align*}
|
||||
\end{itemize}
|
||||
|
||||
{\footnotesize
|
||||
\textbf{Example}: Let $Y \in \{1,2,3\}$ be the clusters, we'd like to assign $X \in \{0,1\}^5$. We're using \textit{Soft Expectation Maximization} to find the cluster assignments $\gamma_j(x_i)$. We assume $\P[X_i = 1 \sep Y]$ follows a $\text{Ber}(p)$ distribution.
|
||||
|
||||
We'd like to find $\gamma_3(x_2) = \P[Y=3 \sep X=x_2]$. The E-step is:
|
||||
$$
|
||||
\P[Y=3 \sep X=x_2] = \frac{\P[X=x_2\sep Y=3]\cdot \overbrace{\P[Y=3]}^{w_3}}{\sum_{i=1}^{3}\P[X=x_2\sep Y=i]\cdot\P[Y=i]}
|
||||
$$
|
||||
For the M-step, we update the assignments via MLE. For $\text{Ber}(p)$, this is the \textit{sample mean}.
|
||||
$$
|
||||
\P[X_i=1 \sep Y=1] = \frac{\sum_{k=1}^{3}\mathbb{I}_{\{x_k^i=1\}}\cdot\P[Y=1\sep X=x_k]}{\sum_{k=1}^3\P[Y=1\sep X=x_k]}
|
||||
$$
|
||||
Where we use $x\in \R^5$
|
||||
}
|
||||
|
||||
\subsubsection{Convergence}
|
||||
|
||||
EM does not generally guarantee global convergence.
|
||||
|
||||
Reference in New Issue
Block a user