//unsigned char buffer[STATIC_ALLOCATOR_SIZE];
//StaticAllocator myAlloc(buffer, STATIC_ALLOCATOR_SIZE);
int main() {
//Alloc::init(&myAlloc);
// DummyAgent agent;
QLearningEGreedyPolicy egreedy(0.1f);
NeuralNetwork net(DIM_OBSERVATIONS + DIM_ACTIONS, N_HIDDEN, 1, 0.1f);
QLearningAgent agent(&net, DIM_OBSERVATIONS, DIM_ACTIONS, N_ACTIONS,
1.0f, 0.1f, &egreedy, false); // lambda = 1.0 => no history
LibMapperEnvironment env;
RLQualia qualia(&agent, &env);
qualia.init();
qualia.start();
for (;;) {
// for (int i=0; i<10; i++) {
qualia.step();
#if is_computer()
printf("Current agent action: %d\n", agent.currentAction.conflated());
printf("Current environment observation: %f %f\n", (double)env.currentObservation.observations[0], (double)env.currentObservation.observations[1]);
#endif
}
// if (myAlloc.nLeaks)
// printf("WARNING: Static Allocator has leaks: %d\n", myAlloc.nLeaks);
return 0;
}
void testMem() {
DummyAgent agent;
DummyEnvironment env;
Qualia qualia(&agent, &env);
qualia.init();
for (int i=0; i<10; i++)
qualia.episode(10);
}
void testDummy() {
DummyAgent agent;
DummyEnvironment env;
Qualia qualia(&agent, &env);
qualia.init();
for (int i=0; i<10; i++) {
qualia.episode(10);
Q_MESSAGE("Current agent action: %d\n", agent.currentAction.conflated());
Q_MESSAGE("Current environment observation: %f %f\n", (double)env.currentObservation.observations[0], (double)env.currentObservation.observations[1]);
}
}
int main(int argc, char** argv) {
srand(RANDOM_SEED);
NeuralNetwork net(GLOW_ENVIRONMENT_OBSERVATIONS_DIM + GLOW_AGENT_ACTIONS_DIM, N_HIDDEN, 1, LEARNING_RATE);
//QLearningEGreedyPolicy policy(EPSILON);
QLearningSoftmaxPolicy policy(0.5f, EPSILON);
QLearningAgent agent(&net,
GLOW_ENVIRONMENT_OBSERVATIONS_DIM, GLOW_AGENT_ACTIONS_DIM, GLOW_AGENT_N_ACTIONS,
1.0f, 0.1f, &policy, false); // lambda = 1.0 => no history
SimpleGlowEnvironment env(argv[1], LED_OUT, PHOTO_AIN);
// BigDummyReward rew;
// DummyRewardEnvironment env(DUMMY_ENVIRONMENT_OBSERVATIONS_DIM, &rew);
RLQualia qualia(&agent, &env);
printf("Starting...\n");
qualia.init();
for (int i=0; i<10; i++) {
printf("# %d ", qualia.nEpisodes);
qualia.episode(1000);
#if is_computer()
printf("Mean reward: %f (%f / %d)\n", (double) qualia.totalReward / qualia.nSteps, qualia.totalReward, qualia.nSteps);
// printf("Current agent action: [%d %d] = %d\n", agent.currentAction[0], agent.currentAction[1], agent.currentAction.conflated());
// printf("Current environment observation: [%f %f] => %f\n", env.currentObservation[0], env.currentObservation[1], env.currentObservation.reward);
#endif
}
// Put epsilon on ice.
printf("Final episode (without random moves)\n");
((QLearningEGreedyPolicy *)agent.policy)->epsilon = 0;
qualia.episode(1000);
#if is_computer()
printf("Mean reward: %f (%f/%d)\n", (double) qualia.totalReward / qualia.nSteps, qualia.totalReward, qualia.nSteps);
printf("Current agent action: [%d] = %d\n", agent.currentAction[0], agent.currentAction.conflated());
// printf("Current environment observation: [%f] => %f\n", env.currentObservation[0], env.currentObservation.reward);
#endif
return 0;
}
int main() {
DummyAgent agent(DIM_ACTIONS, N_ACTIONS);
MapperBasicEnvironment env(DIM_OBSERVATIONS, DIM_ACTIONS, "agent", PEER_DEVICE_NAME, true, 9000);
Qualia qualia(&agent, &env);
qualia.init();
qualia.start();
for (;;) {
// for (int i=0; i<10; i++) {
qualia.step();
#if is_computer()
printf("Current agent action: %d\n", (int)agent.currentAction.conflated());
printf("Current environment observation: %f %f\n", (double)env.currentObservation.observations[0], (double)env.currentObservation.observations[1]);
#endif
}
// if (myAlloc.nLeaks)
// printf("WARNING: Static Allocator has leaks: %d\n", myAlloc.nLeaks);
return 0;
}
void testQLearning(Environment& env, QLearningAgent& agent) {
RLQualia qualia(&agent, &env);
Q_MESSAGE("Starting...");
#if is_computer()
Q_MESSAGE("(this is a computer)");
#endif
qualia.init();
agent.isLearning = false;
qualia.episode(1000);
Q_MESSAGE("First episode: no learning");
Q_MESSAGE("Mean reward: %f (%f / %d)", (double) qualia.totalReward / qualia.nSteps, qualia.totalReward, qualia.nSteps);
// Q_MESSAGE("Current agent action: [%d %d] = %d", agent.currentAction[0], agent.currentAction[1], agent.currentAction.conflated());
// Q_MESSAGE("Current environment observation: [%f %f] => %f", env.currentObservation[0], env.currentObservation[1], env.currentObservation.reward);
for (int i=0; i<10; i++) {
qualia.episode(1000);
agent.isLearning = true;
Q_MESSAGE("# %d ", qualia.nEpisodes);
Q_MESSAGE("Mean reward: %f (%f / %d)", (double) qualia.totalReward / qualia.nSteps, qualia.totalReward, qualia.nSteps);
// Q_MESSAGE("Current agent action: [%d %d] = %d", agent.currentAction[0], agent.currentAction[1], agent.currentAction.conflated());
// Q_MESSAGE("Current environment observation: [%f %f] => %f", env.currentObservation[0], env.currentObservation[1], env.currentObservation.reward);
}
// Put epsilon on ice.
((QLearningEGreedyPolicy *)agent.policy)->epsilon = 0;
qualia.episode(1000);
Q_MESSAGE("Final episode (without random moves)");
Q_MESSAGE("Mean reward: %f (%f/%d)", (double) qualia.totalReward / qualia.nSteps, qualia.totalReward, qualia.nSteps);
Q_MESSAGE("Current agent action: [%d] = %d", agent.currentAction[0], agent.currentAction.conflated());
// Q_MESSAGE("Current environment observation: [%f] => %f", env.currentObservation[0], env.currentObservation.reward);
}
