protected void Reset()
{
// reset the agent"s policy and value function
Q = Tembo.ArrayOfZeros(NS * NA);
if (Options.QInitVal != 0)
{
Tembo.SetConst(Q, Options.QInitVal);
}
P = Tembo.ArrayOfZeros(NS * NA);
E = Tembo.ArrayOfZeros(NS * NA);
// model/planning vars
EnvModelS = Tembo.ArrayOfZeros(NS * NA);
Tembo.SetConst(EnvModelS, -1); // init to -1 so we can test if we saw the state before
EnvModelR = Tembo.ArrayOfZeros(NS * NA);
SaSeen = new double[] { };
PQ = Tembo.ArrayOfZeros(NS * NA);
// initialize uniform random policy
for (var s = 0; s < NS; s++)
{
var poss = AllowedActions(s);
for (var i = 0; i < poss.Length; i++)
{
P[poss[i] * NS + s] = 1.0 / poss.Length;
}
}
// agent memory, needed for streaming updates
// (s0,a0,r0,s1,a1,r1,...)
r0 = 999999999;
s0 = 999999999;
s1 = 999999999;
a0 = 999999999;
a1 = 999999999;
}
