2.3. The general model and its known weaknesses
We allowed the machine learners to use both the same
and an alternate set of features compared to the expert BN.
SpeciWcally, the initial set of factors were the basic factors
in the expert model, plus all the other registered Spurs’
players (as playing or not playing) rather than just the four
‘special’ players in the expert BN minus the playing position
of Wilson. The particular values for Opposition quality
in each game were the same as those used by the expert BN.
During a game players can be injured, substituted, be
sent oV, or have their playing positions changed. The solution
chosen to deal with these issues was to use the information
about only those players who started the game.
Similarly Wilson’s playing position could change during
the course of the match, only his initial playing position
was considered.
In general terms this problem is not particularly easy
from a machine learning perspective. There is not much
data to go on. We have the results of two seasons’ games, a
total of 76 matches and for the general model a total of 30
attributes, (28 players, venue, and opponent quality). There
were changes to the Spurs’ squad during this period. The
simple convention of a player either playing or not was
chosen to avoid having missing data entries with regards to
2.3. The general model and its known weaknesses
We allowed the machine learners to use both the same
and an alternate set of features compared to the expert BN.
SpeciWcally, the initial set of factors were the basic factors
in the expert model, plus all the other registered Spurs’
players (as playing or not playing) rather than just the four
‘special’ players in the expert BN minus the playing position
of Wilson. The particular values for Opposition quality
in each game were the same as those used by the expert BN.
During a game players can be injured, substituted, be
sent oV, or have their playing positions changed. The solution
chosen to deal with these issues was to use the information
about only those players who started the game.
Similarly Wilson’s playing position could change during
the course of the match, only his initial playing position
was considered.
In general terms this problem is not particularly easy
from a machine learning perspective. There is not much
data to go on. We have the results of two seasons’ games, a
total of 76 matches and for the general model a total of 30
attributes, (28 players, venue, and opponent quality). There
were changes to the Spurs’ squad during this period. The
simple convention of a player either playing or not was
chosen to avoid having missing data entries with regards to
การแปล กรุณารอสักครู่..