update

diff --git a/week3/Makefile b/week3/Makefile
index feeddb6..b7e1f69 100644 (file)
--- a/week3/Makefile
+++ b/week3/Makefile
@@ -5,7 +5,8 @@ PDFLATEX:=pdflatex $(PDFLATEXFLAGS)
 all: $(addsuffix .pdf,$(DOCUMENTS))
 
 %.pdf: %.tex
-       pdflatex $<
+       $(PDFLATEX) $<
+       $(PDFLATEX) $<
 
 clean:
        $(RM) -v $(addprefix $(DOCUMENTS),.pdf .log .aux)

diff --git a/week3/eml.tex b/week3/eml.tex
index 781f2fe..e44cd40 100644 (file)
--- a/week3/eml.tex
+++ b/week3/eml.tex
@@ -29,10 +29,10 @@ Using ten-fold cross validation decreases the percentage.
                \toprule
                Method & Correctly classified & Root relative squared error\\
                \midrule
-               NaiveBayes & $96.6449\%$ & $35.7222\%$\\
-               NaiveBayes (10FCF) & $96.4352\%$ & $37.1926\%$\\
-               J48 & $96.6449\%$ & $34.9136\%$\\
-               J48 (10FCF) & $96.4352\%$ & $36.5122\%$\\
+               \emph{NaiveBayes} & $96.6449\%$ & $35.7222\%$\\
+               \emph{NaiveBayes (10FCF)} & $96.4352\%$ & $37.1926\%$\\
+               \emph{J48} & $96.6449\%$ & $34.9136\%$\\
+               \emph{J48 (10FCF)} & $96.4352\%$ & $36.5122\%$\\
                \bottomrule
        \end{tabular}
        \caption{Results for \texttt{P1D} and \texttt{FD}\label{t1}}
@@ -65,7 +65,7 @@ Using ten-fold cross validation decreases the percentage.
                & & & & \checkmark{} & $88.4436\%$\\
                \bottomrule
        \end{tabular}
-       \caption{NaiveBayes on all sensible combinations\label{t2}}
+       \caption{\emph{NaiveBayes} on all sensible combinations\label{t2}}
 \end{table}
 
 \subsection*{Chapter 6: Exercises}
@@ -78,8 +78,19 @@ Using ten-fold cross validation decreases the percentage.
                sequence, i.e.\ both the positions more than $n$ back and the following
                positions.}
 
+               A probability is always based on the most probably preceding sequence,
+               however there are no backpointers to all states. Thus the probability
+               is not based on all possible previous paths and you can only recover
+               the most likely path. The following states are also of an influence
+               since when the path does not belong to the most likely sequence it will
+               not be connected to the final path via a backpointer and will be lost.
+
        \item\emph{Explain in your own words (at most 50) how the EM algorithm
                works. I don't mean the mathematics, but the underlying concept.}
 
+               The \emph{Expectation-Maximization}-algorithm (EM) searches for the
+               settings of parameters where the likelihood is (locally) optimal. 
+               The algorithm usually takes the derivative of the likelihood function
+               to get the maximum value.
 \end{itemize}
 \end{document}