Browse files

journal: model testing

  • Loading branch information...
1 parent d00c5f6 commit 5696a9aa79e74a2b6065e492f200c250b6aba743 @spyked spyked committed May 17, 2012
Showing with 16 additions and 1 deletion.
  1. +16 −1 journal/journal.tex
@@ -290,7 +290,22 @@ \subsection*{Model testing}
After obtaining a model, we need to test it on a separate data set. We do this
using mostly the same infrastructure, as described in Section
+\ref{sec:architecture}. The generation and conversion into a libSVM-readable
+format are done in the same fashion as the training phase. The testing is
+however done using \texttt{svm-predict} (also provided by libSVM), which takes
+as input a dataset and a previously-generated model and outputs the accuracy --
+provided that the classes of the newly-generated dataset are known -- of the
+prediction as well as a file that contains, in order, the class of each example
+in the dataset.
+We automated the process of generation-conversion-prediction using the
+\texttt{predict.bash} script. Also, to provide a better readability of the
+output file generated by \texttt{svm-predict} we developed a tool called
+\texttt{svm2out} that takes as input a file containing classes and outputs the
+actual predicted letters.
+The testing phase currently randomly generates a number of 100 examples used
+for evaluation. The metric used for evaluation is accuracy.
\subsection*{Finding a suitable model}

0 comments on commit 5696a9a

Please sign in to comment.