+In order to be allow learnlib to learn the TCP model it was necessary to have a deterministic model.\r
+We accomplished this by modifying the adapter so it can reach a \emph{ERROR} or \emph{CLOSED} state. In these states all inputs are discarded and a default output is returned.\r
+In the case of a state where an input results in a non-deterministic output we jump to the \emph{ERROR} state for additional this given input. When the connection is successfully closed using a \emph{FIN} packet we move the adapter to the \emph{CLOSED} state.\r
+\r
+We divided the input alphabet into three sets, this way we can control the size of the model learned by learnlib.\r
+\r
+\begin{longtable}{|c|l|}\r
+ \caption{Different input alphabets used during learning.} \\\hline\r
+ Alphabet & Inputs \\\hline \hline\r
+ small & SYN, ACK \\\hline\r
+ partial & SYN, ACK, DATA \\\hline\r
+ full & SYN, ACK, DATA, RST, FIN \\\hline\r
+\end{longtable}\r
+\r
+Just as in our previous assignment the \emph{DATA} packet is actually a \emph{ACK} with an user data payload and the \emph{push} flag set. \r
+These input alphabets will influence the size of the model produced. \emph{small} will result in a 2 state model, \emph{partial} will be the full model without the \emph{CLOSED} state and \emph{full} should result in the full model as used in the previous assignment.\r
+\r
+\paragraph{Model learned with small input alphabet}\r
+\includegraphics{model.small.LStar.rand.eps}\r
+\r
+\r
+\paragraph{Model learned with partial input alphabet}\r
+\includegraphics{model.partial.LStar.rand.eps}\r
+\r
+\paragraph{Model learned with full input alphabet}\r
+\includegraphics{model.full.LStar.rand.eps}
\ No newline at end of file
repositories / tt2015.git / blobdiff