In this section, we will take you through a short tutorial, to help you get
acquainted with the main features of MTK. If you are already familiar with them,
and are looking for detailed information on a specific plugin, we recommend you
jump right to section , were we describe, individually, each of
the available plugins. Our goal, here, will be to build a hidden Markov model to
adequately represent a coin-tossing game, described below, and use it to
forecast its future outcomes.
Suppose you are in a casino and, while walking through the main floor, you
notice that there is a new strange game available, called Thank Paty the
Parrot!. Curious, you walk up to the table, which has a green parrot standing
on top of it, and ask the dealer about the game. Its rules are very simple.
There are two coins inside a small basket, over the table, in front of the
parrot. In each round, the parrot randomly chooses one, picking it up with his
beak, and hands it over to the dealer. The dealer, then, flips the coin into the
air. If it lands showing tails, you win dollar for every dollar bet,
doubling your money. Otherwise, if it lands showing heads, the casino keeps all
the money you bet. After the bets are payed, the dealer puts the coin back into
the basket, shuffles it, and places it, again, in front of the parrot.
When asked about the coins, the dealer says that, despite being visually identical, they have different biases. One has a significant greater probability of showing heads, and the other has a significant probability of showing tails. So your chances of winning are conditioned on the parrots choice, thus the name Thank Paty the Parrot!. Paty, as you, can't tell the difference between both coins, but knows that every time he chooses one, he gets a nice pat on the head.
After hearing the explanation, you remember having read, just last week, a paper on hidden Markov models (HMM) [#!rabiner!#], and how they can be applied to situations just like the one described to you. Feeling that, finally, all that studying might pay off, you convince yourself that this is the game that's going to get you some money!
But before playing, you have to build your HMM model. To do this, you must first
collect some observations, that will be used to estimate the parameters of the
model. So you decide to observe others playing, and take notes on the outcomes
of the coins. Your observations are recorded in the file trace.txt11.1 in which a represents tails, and a
represents heads. After
collecting
observations11.2, you head home and, using MTK, start building
your HMM model.