From 3befc8bf713692d786de259e9782345858222580 Mon Sep 17 00:00:00 2001
From: Piers Williams <solar_1992@hotmail.co.uk>
Date: Mon, 21 May 2018 10:22:40 +0100
Subject: [PATCH] Update
---
slides/content/ai.tex | 26 ++++++++++++++++++++++----
1 file changed, 22 insertions(+), 4 deletions(-)
diff --git a/slides/content/ai.tex b/slides/content/ai.tex
index 23413c0..a209204 100644
--- a/slides/content/ai.tex
+++ b/slides/content/ai.tex
@@ -320,8 +320,8 @@ List<UUID> entities
\begin{frame}{GA for Simple Game}
\begin{itemize}[<+->]
\item Given a time budget
- \item Spend it evolving ``plans''
- \item Plan is a sequence of possible actions
+ \item Spend it \textbf{evolving} ``plans''
+ \item Plan is a \textbf{sequence} of possible actions
\item Simulate the plan and evaluate resultant state for score
\item Sound like a possible GA?
\item This is called a Rolling Horizon Evolutionary Algorithm
@@ -330,8 +330,9 @@ List<UUID> entities
\begin{frame}{Boosting the horizon}
\begin{itemize}[<+->]
- \item RHEA is quite jerky in games
- \item Macro Actions can solve this
+ \item Length of the sequence increases parameters for GA
+ \item RHEA is quite \textbf{jerky} in games
+ \item Macro Actions can sometimes solve this
\item Locking the agent to consider each move $N$ times
\item Means you can think for $N$ turns
\item Works great in real-time engines like PTSP
@@ -407,6 +408,23 @@ List<UUID> entities
\end{itemize}
\end{frame}
+\begin{frame}{Problems}
+ \begin{tabularx}{6cm}{X | r r}
+ Controller & ops/min & Error \\ \hline
+ random & 512.840 & 13.846 \\
+ noop & 509.314 & 5.730 \\
+ RangedRush & 273.362 & 2.345 \\
+ MixedRush & 284.875 & 7.453 \\
+ MedievalGA & 3.538 & 2.286 \\
+ \end{tabularx}
+
+\begin{itemize}
+ \item An ``op'' is building a full game and playing it to the end
+ \item MedievalGA is the VLMAGA
+ \item I wish that were an error for MedievalGA
+\end{itemize}
+\end{frame}
+
\begin{comment}
\section{Monte-Carlo Tree Search}
\begin{frame}[fragile]{Monte-Carlo Tree Search}
--
GitLab