/// <summary>
/// Try to get the length of arguments of a method
/// </summary>
/// <param name="pyObject">Object representing a method</param>
/// <param name="length">Lenght of arguments</param>
/// <returns>True if pyObject is a method</returns>
private static bool TryGetArgLength(PyObject pyObject, out long length)
{
using (Py.GIL())
{
var inspect = lazyInspect.Value;
if (inspect.isfunction(pyObject))
{
var args = inspect.getargspec(pyObject).args as PyObject;
var pyList = new PyList(args);
length = pyList.Length();
pyList.Dispose();
args.Dispose();
return(true);
}
if (inspect.ismethod(pyObject))
{
var args = inspect.getargspec(pyObject).args as PyObject;
var pyList = new PyList(args);
length = pyList.Length() - 1;
pyList.Dispose();
args.Dispose();
return(true);
}
}
length = 0;
return(false);
}
public void TestEmptyCtor()
{
var s = new PyList();
Assert.IsInstanceOf(typeof(PyList), s);
Assert.AreEqual(0, s.Length());
}
public static void LoadDefaultArgs()
{
using (new PythonEngine())
using (var argv = new PyList(Runtime.Runtime.PySys_GetObject("argv")))
{
Assert.AreNotEqual(0, argv.Length());
}
}
public void TestPyObjectCtor()
{
var a = new PyList();
var s = new PyList(a);
Assert.IsInstanceOf(typeof(PyList), s);
Assert.AreEqual(0, s.Length());
}
public static void LoadDefaultArgs()
{
using (new PythonEngine())
{
// PySys_GetObject get a borrowed reference, it just no need for dispose
var argv = new PyList(Runtime.Runtime.PySys_GetObject("argv"));
Assert.AreNotEqual(0, argv.Length());
}
}
public void TestInsert()
{
var ai = new PyObject[] { new PyInt(3), new PyInt(2), new PyInt(1) };
var s = new PyList(ai);
s.Insert(0, new PyInt(4));
Assert.AreEqual(4, s.Length());
Assert.AreEqual("4", s[0].ToString());
}
public void ToPyList()
{
var list = new PyList();
list.Append("hello".ToPython());
list.Append("world".ToPython());
var back = list.ToPython().As <PyList>();
Assert.AreEqual(list.Length(), back.Length());
}
public void TestPyObjectArrayCtor()
{
var ai = new PyObject[] { new PyInt(3), new PyInt(2), new PyInt(1) };
var s = new PyList(ai);
Assert.IsInstanceOf(typeof(PyList), s);
Assert.AreEqual(3, s.Length());
Assert.AreEqual("3", s[0].ToString());
Assert.AreEqual("2", s[1].ToString());
Assert.AreEqual("1", s[2].ToString());
}
public void TestSort()
{
var ai = new PyObject[] { new PyInt(3), new PyInt(1), new PyInt(2) };
var s = new PyList(ai);
s.Sort();
Assert.AreEqual(3, s.Length());
Assert.AreEqual("1", s[0].ToString());
Assert.AreEqual("2", s[1].ToString());
Assert.AreEqual("3", s[2].ToString());
}
private static double[,] ConvertSimilarityPythonListToArray(PyList similarityList)
{
var length = similarityList.Length();
var similarities = new double[length, length];
for (int i = 0; i < length; i++)
{
var columnRows = PyList.AsList(similarityList[i]);
var row = (double[])columnRows.AsManagedObject(typeof(double[]));
for (int j = 0; j < length; j++)
{
similarities[i, j] = row[j];
}
}
return(similarities);
}
private void DQN(object sender, RoutedEventArgs e)
{
bool side = true; //own side
int numGames = 5000;
using (Py.GIL())
{
dynamic dqn = Py.Import("simple_dqn.dqn");
dynamic Agent = dqn.Agent;
Console.WriteLine("Done importing");
PyInt width = new PyInt(458);
dynamic state_size = new PyTuple(new PyObject[] { width });
dynamic agent = Agent(Py.kw("state_size", state_size), Py.kw("discount", 0.95));
Console.WriteLine(agent.discount);
for (int i = 0; i < numGames; i++)
{
PyFloat epsilon = new PyFloat(Math.Max(0.1, 1 - (double)(i + 1) * 4 / numGames));
agent.epsilon = epsilon;
if (!isInit)
{
Init(true, i);
}
Playfield state = GameManager.Instance.mPlayfield;
int result = state.getGameResult();
float total_cost = 0.0f;
float total_reward = 0.0f;
float reward = 0.0f;
float cost = 0.0f;
int num_move = 0;
int end_turn_count = 0;
agent.new_episode();
GreedyPlayer greedyPlayer = (GreedyPlayer)GameManager.Instance.playerFirst;
while (result == -1)
{
Macro moveTodo = null;
reward = 0.0f;
cost = 0.0f;
if (state.isOwnTurn)
{
List <Macro> macroMoveList = greedyPlayer.getMacros(state);
macroMoveList.Add(null);
dynamic cur_state_vector = DNNEval.Instance.parsePlayfield(state, side);
List <Playfield> stateList = new List <Playfield>();
PyList state_vector_list = new PyList();
foreach (Macro macro in macroMoveList)
{
Playfield nextState = new Playfield(state);
if (macro != null)
{
nextState.doMacroAction(macro);
stateList.Add(nextState);
reward = nextState.observerMoveReward();
}
else
{
nextState.endTurn(false, false);
nextState.drawTurnStartCard();
//simulateEnemyTurn(nextState, false, true);
}
dynamic state_vector = DNNEval.Instance.parsePlayfield(nextState, side);
state_vector_list.Append(state_vector);
}
int action_index = agent.act(cur_state_vector, state_vector_list);
Helpfunctions.Instance.logg("Length:" + state_vector_list.Length());
moveTodo = macroMoveList[action_index];
if (moveTodo != null)
{
Helpfunctions.Instance.logg("Player 1 ##########Move##########");
//moveTodo.print();
state.doMacroAction(moveTodo);
reward = state.observerMoveReward();
num_move += 1;
cost = agent.observe(reward);
Helpfunctions.Instance.logg("action_index = " + action_index + "/" + (macroMoveList.Count - 1) + "reward = " + reward);
}
else
{
if (state.playerFirst.mana == state.playerFirst.ownMaxMana)
{
end_turn_count++;
}
Helpfunctions.Instance.logg("Player 1 ##########Move##########");
Helpfunctions.Instance.logg("Player 1 end turn");
state.endTurn(false, false);
state.drawTurnStartCard();
dynamic end_state_vector = DNNEval.Instance.parsePlayfield(state, side);
agent.update_transition(end_state_vector);
reward = 0.0f;
num_move += 1;
cost = agent.observe(reward);
Helpfunctions.Instance.logg("action_index = " + action_index + "/" + (macroMoveList.Count - 1) + "reward = " + reward);
total_reward += simulateEnemyTurn(state, false, false, agent);
Helpfunctions.Instance.logg("Player 2 end turn");
}
//state.printBoard();
total_cost += cost;
total_reward += reward;
result = state.getGameResult();
}
else
{
int debug = 1;
}
}
if (result == 0)
{
int debug = 1;
}
isInit = false;
Helpfunctions.Instance.logg("last reward: " + reward);
Helpfunctions.Instance.logg("total reward: " + total_reward);
Helpfunctions.Instance.logg("avg cost: " + total_cost / num_move);
Helpfunctions.Instance.WriteResultToFile(@"\learning_log.txt", "No." + i + " total/avg cost: " + total_reward + ", " + total_cost / num_move + ", end turn: " + (float)end_turn_count / num_move);
}
}
//for e in xrange(num_episodes):
// observation = env.reset()
// done = False
// agent.new_episode()
// total_cost = 0.0
// total_reward = 0.0
// frame = 0
// while not done:
// frame += 1
// #env.render()
// action, values = agent.act(observation)
// #action = env.action_space.sample()
// observation, reward, done, info = env.step(action)
// print reward
// total_cost += agent.observe(reward)
// total_reward += reward
// print "total reward", total_reward
// print "mean cost", total_cost/frame
}
private void DQNTurn(object sender, RoutedEventArgs e)
{
bool side = true; //own side
int numGames = 500;
using (Py.GIL())
{
dynamic dqn = Py.Import("simple_dqn.dqn");
dynamic Agent = dqn.Agent;
Console.WriteLine("Done importing");
PyInt width = new PyInt(458);
dynamic state_size = new PyTuple(new PyObject[] { width });
dynamic agent = Agent(Py.kw("state_size", state_size), Py.kw("discount", 0.95));
Console.WriteLine(agent.discount);
for (int i = 0; i < numGames; i++)
{
if (!isInit)
{
Init(true, i);
}
Playfield state = GameManager.Instance.mPlayfield;
int result = state.getGameResult();
float total_cost = 0.0f;
float total_reward = 0.0f;
float reward = 0.0f;
float cost = 0.0f;
int num_move = 0;
int end_turn_count = 0;
agent.new_episode();
while (result == -1)
{
Action moveTodo = null;
reward = 0.0f;
cost = 0.0f;
if (state.isOwnTurn)
{
List <Action> moveList = Movegenerator.Instance.getMoveList(state, false, true, true, 0.0);
moveList.Add(null);
dynamic cur_state_vector = DNNEval.Instance.parsePlayfield(state, side);
List <Playfield> stateList = new List <Playfield>();
PyList state_vector_list = new PyList();
foreach (Action action in moveList)
{
Playfield nextState = new Playfield(state);
if (action != null)
{
nextState.doAction(action);
stateList.Add(nextState);
reward = nextState.observerMoveReward();
}
else
{
nextState.endTurn(false, false);
nextState.drawTurnStartCard();
//simulateEnemyTurn(nextState, false, true, agent);
}
dynamic state_vector = DNNEval.Instance.parsePlayfield(nextState, side);
state_vector_list.Append(state_vector);
}
int action_index = agent.act(cur_state_vector, state_vector_list);
Helpfunctions.Instance.logg("Length:" + state_vector_list.Length());
moveTodo = moveList[action_index];
if (moveTodo != null)
{
Helpfunctions.Instance.logg("Player 1 ##########Move##########");
moveTodo.print();
state.doAction(moveTodo);
reward = state.observerMoveReward();
}
else
{
if (state.playerFirst.mana == state.playerFirst.ownMaxMana)
{
end_turn_count++;
}
Helpfunctions.Instance.logg("Player 1 ##########Move##########");
Helpfunctions.Instance.logg("Player 1 end turn");
state.endTurn(false, false);
state.drawTurnStartCard();
reward = 0.0f;
total_reward += simulateEnemyTurn(state, false, false, agent);
Helpfunctions.Instance.logg("Player 2 end turn");
}
//state.printBoard();
num_move += 1;
cost = agent.observe(reward);
Helpfunctions.Instance.logg("reward = " + reward);
total_cost += cost;
total_reward += reward;
result = state.getGameResult();
}
else
{
int debug = 1;
}
}
if (result == 0)
{
int debug = 1;
}
isInit = false;
Helpfunctions.Instance.logg("last reward: " + reward);
Helpfunctions.Instance.logg("total reward: " + total_reward);
Helpfunctions.Instance.logg("avg cost: " + total_cost / num_move);
Helpfunctions.Instance.WriteResultToFile(@"\learning_log.txt", "No." + i + " total/avg cost: " + total_reward + ", " + total_cost / num_move + ", end turn: " + (float)end_turn_count / num_move);
}
}
}
