Game Development Reference
In-Depth Information
of our method. Somehow, we had to decide about the right values for various action
weights and thresholds so the method could correctly evaluate the featured tags for
a minimum number of player's actions. Therefore, we decided to firstly set them
by hand, collect the dataset of the game logs, and then, using simulations, tune the
parameter values for best performance. Also, as a part of our method, we created
the combinations of parameters with the input values (computed from game logs).
These expressions were firm, and have not undergone changes.
We have identified following parameters, input values, constants (parameters that
have not been a subject of training) and variables that feature in our tag validation
mechanism:
support —a tag validity support variable itself. At the start of the process, it is set
to zero.
Δ
support —a change of the support value.
ε upper —a threshold of the support value, a positive number and a constant.
If support reaches this value, the tag is confirmed as correct.
ε lower —a threshold of the support value, a negative number and a constant.
If support reaches this value, the tag is rejected as incorrect.
nbr —A number of choices available at the time of the player action.
attm —A number of player attempts in this question, including the current action.
con f —Aconfidence of the player. Anumber between 0 and 1, where 1 corresponds
to the highest possible bet.
ans c —A weight of the support change in case of “correct” choice was picked.
ans w —A weight of the support change in case of incorrect choice was picked.
tag c —A weight of the support change in case when player selects a possibly
correct tag in a wrong choice.
tag w —A weight of the support change in case when player rules-out a incorrect
tag in the “correct” choice.
buy —A weight of the support change when player buys an additional tag (the
change affects the other tags present in the choice).
The latter five are the parameters to be trained. The following are the situations,
where the support value is altered due to the player's action:
The player chooses a “correct” choice
The positive support change. It is influenced by player's confidence and number of
other choices (the higher, the harder is the question). Also, the positive impact is
exponentially decreased when the choice was selected after more attempts.
ans c
con f
nbr
Δ
support
=
2 attm
The player chooses an incorrect choice
The negative support change. Unlike in the previous case, the impact is decreased
by the number of other choices.
Search WWH ::




Custom Search