def main():
numActionDirections = 8
actionSpace = [(np.cos(directionId * 2 * math.pi / numActionDirections),
np.sin(directionId * 2 * math.pi / numActionDirections))
for directionId in range(numActionDirections)]
actionSpace.append((0, 0))
stateDim = 100
actionDim = len(actionSpace)
I = 64
J = 60
for K in [0.001, 0.0001, 0.00001]:
paramSet = {
'INITIAL_EPSILON': 0.4,
'FINAL_EPSILON': 0.01,
'GAMMA': 0.99,
'REPLAY_SIZE': 10000,
'BATCH_SIZE': I,
'REPLACE_TARGET_FREQ': 20,
'HIDDEN_LAYER_WIDTH': J,
'LR': K,
}
dqn = dqnModel(stateDim, actionDim, paramSet)
trainTask(dqn, actionSpace)
print("# batch size:{},width:{},lr:{}".format(I, J, K))
print("#-----------------------------------")
def main():
numActionDirections = 8
actionSpace = [(np.cos(directionId * 2 * math.pi / numActionDirections),
np.sin(directionId * 2 * math.pi / numActionDirections))
for directionId in range(numActionDirections)]
stateDim = 100
actionDim = len(actionSpace)
print(actionDim)
I=128
J=60
K=0.0001
paramSet = {
'INITIAL_EPSILON': 0.4,
'FINAL_EPSILON': 0.01,
'GAMMA': 0.99,
'REPLAY_SIZE': 10000,
'BATCH_SIZE': I,
'REPLACE_TARGET_FREQ': 20,
'HIDDEN_LAYER_WIDTH': J,
'LR': K,
}
dqn = dqnModel(stateDim, actionDim, paramSet)
trainTask(dqn,actionSpace)