We begin by looking more closely at some simple algorithms for estimating the
value of actions and for using the estimates to make action-selection decisions.
In this chapter, we denote the true (actual) value of action a as , and the
estimated value after t plays as
. Recall that the true value
of an action is the mean reward given that the action is selected. One natural way to
estimate this is by averaging the rewards actually received when the
action was selected. In other words, if, after t decisions, action a has been
chosen
times, yielding rewards
, then its value is
estimated to be
If , then we define
instead as some default value, e.g.,
.
As
, by the law of large numbers
converges to
. We call this the sample-average method for estimating action values
because each estimate is a simple average of the sample of relevant rewards.
Of course this is just one way to estimate action values, not necessarily the best
one. Nevertheless, for now let us stay with this simple estimation algorithm and
turn to the question of how the estimates might be used to select actions.
The simplest action selection rule is to select the action (or one of the
actions) with highest estimated action value, i.e., to select on the tth
play one of the greedy actions,
, for which
. This method
always exploits current knowledge to maximize immediate reward; it spends
no time at all sampling apparently inferior actions to see if they might
really be better. A simple alternative is to behave greedily most of the
time, but every once in a while, say with small probability
, instead
select an action at random, uniformly, independently of the action-value
estimates. We call methods using this near-greedy action selection rule
-greedy methods. An advantage of these methods is that in the limit
as the number of plays increases, every action will be sampled an
infinite number of times, guaranteeing that
for all
a, and thus ensuring that all the
converge to
. This of course
implies that the probability of selecting the optimal action converges to
,
i.e., to near certainty. These are just
asymptotic guarantees, however, and say little about the practical effectiveness
of the methods.
To roughly assess the relative effectiveness of the greedy and
-greedy methods, we
compared them numerically on a suite of test problems. This is a set of 2000 randomly
generated n
-armed bandit tasks with n=10. For each action, a, the rewards were
selected from a normal (Gaussian) probability distribution with mean
and
variance 1. The 2000 n
-armed bandit tasks were generated by
re-selecting the
2000 times, each according to a normal distribution with
mean
0 and variance 1. Averaging over tasks, we can plot the performance and
behavior of various methods as they improve with experience over 1000 plays, as in
Figure 2.1. We call this suite of test tasks the
10-armed testbed.
Figure 2.1: Average performance of
-greedy action-value methods on 10-armed testbed.
These data are averages over 2000 tasks. All methods used sample averages as
their action-value estimates.
Figure 2.1 compares a greedy method with two
-greedy methods (
and
), as described above, on the 10-armed
testbed. Both methods formed their action-value estimates using the sample-average
technique. The upper graph shows the increase in expected reward with experience. The
greedy method improves slightly faster than the other methods at the very
beginning, but then levels off at a lower level. It achieves a reward-per-step of
only about 1 compared to the best possible of about 1.55 on this testbed. The greedy
method performs significantly worse in the long run because it often gets stuck
performing suboptimal actions. The lower graph shows that the greedy method found
the optimal action only in approximately one-third of the tasks. In the other
two-thirds, its initial samples of the optimal action were disappointing, and it
never returned to it. The
-greedy methods eventually perform better because they continue to explore, and
continue to improve their chances of recognizing the optimal action. The
method explores more, and usually finds the optimal action earlier, but never
selects it more than 91% of the time. The
method improves more slowly,
but eventually performs better than the
method by both performance
measures. It is also possible to reduce
over time to try to get the best of both
high and low values.
The advantage of
-greedy over greedy methods depends on the problem. For example,
suppose the reward variance had been larger, say 10 instead of 1. With noisier
rewards it takes more exploration to find the optimal action, and
-greedy
methods should fare even better relative to the greedy method.
On the other hand, if the reward variances were zero, then the greedy method would
know the true value of each action after trying it once. In this case the greedy method
might actually perform best because it would soon find the optimal action and then never
explore. But even in the deterministic case, there is a large advantage to exploring if
we weaken some of the other assumptions. For example, suppose the bandit problem were
nonstationary, that is, that the true values of the actions changed over time. In this
case exploration is needed even in the deterministic case to make sure one of the
non-greedy actions has not changed to become better than the greedy one. As we
will see in the next few chapters, effective nonstationarity is the case most commonly
encountered in reinforcement learning. Even if the underlying problem is stationary
and deterministic, the learner faces a set of bandit-like decision problems each
of which changes over time due to the learning process itself. Reinforcement learning
problems have a strong requirement for some sort of
balance between exploration and exploitation.
Exercise .
In the comparison shown in Figure 2.1, which method will perform best in the long run, in terms of cumulative reward and cumulative probability of selecting the best action? How much better will it be?