//Copied from another algorithm state
//We reset some data, so we dont reflect values that aren't true for the new state
//This constructor is called when a new step is being generated, so we transfer some values appropriately.
public AlgorithmState(AlgorithmState set_from)
{
cans_collected = set_from.cans_collected;
episode_rewards = set_from.episode_rewards; //Reward data
total_rewards = set_from.total_rewards;
board_data = new GameBoard(set_from.board_data); //Copy the board
//Increase steps in here
live_qmatrix = new Qmatrix(set_from.live_qmatrix); //Copy the q matrix
//The initial location will be the resulting location of the last step
location_initial = new int[2] {
set_from.location_result[0], set_from.location_result[1]
};
bender_perception_starting = set_from.bender_perception_ending;
//Detect if we reached the limit for this episode
if (live_qmatrix.step_number == Qmatrix.step_limit)
{
StartNewEpisode();
}
else
{
live_qmatrix.step_number++;
}
}
//Called from create_empty_board (after reset), and the constructor
//Just a useful container for resetting some values when we want to start over, but making a new state would have us lose bender's position.
private void InitializeValues()
{
board_data.ClearCans(); //Clear the board for our initial launch(this doesn't remove bender, just cans)
live_qmatrix = new Qmatrix();
location_initial = new int[2] {
0, 0
};
location_result = new int[2] {
0, 0
};
}
public Qmatrix(Qmatrix copy_from)
{
//Copy the q-matrix.
matrix_data = new Dictionary <PerceptionState, ValueSet>();
foreach (var i in copy_from.matrix_data.Keys)
{ //For each list of strings in copy_from.matrix data
//Get a copy of the dictionary at this list of strings
//Should be a deep copy
matrix_data.Add(i, new ValueSet(copy_from.matrix_data[i]));
}
did_we_update = copy_from.did_we_update;
step_number = copy_from.step_number;
episode_number = copy_from.episode_number;
n_current = copy_from.n_current;
y_current = copy_from.y_current;
e_current = copy_from.e_current;
}
