// Implement to manage challenge levels and training set hints
public virtual void IncreaseLevel()
{
CurrentLevel++;
if (CurrentLevel >= NumberOfLevels)
{
// this case happens when the task is finished
if (CurrentLevel > NumberOfLevels)
{
// this case should not happen
Debug.Assert(false);
}
CurrentLevel = NumberOfLevels;
return;
}
// We assume that levels are traversed sequentially.
// Random access of levels would require a change of
// implementation.
CurrentNumberOfAttempts = 0;
CurrentNumberOfSuccesses = 0;
TSHints.Set(TSProgression[CurrentLevel]);
SetHints(TSHints);
MyLog.Writer.WriteLine(MyLogLevel.INFO,
"Next level settings: \n" +
TSHints);
}
public LTMovingTarget(SchoolWorld w)
: base(w)
{
TSHints[TSHintAttributes.DEGREES_OF_FREEDOM] = 2; // Set degrees of freedom to 2: move in 4 directions (1 means move only right-left)
//TSHints[TSHintAttributes.MAX_TARGET_DISTANCE] = 0.3f;
TSHints[DISTANCE_BONUS_COEFFICENT] = 15f; // Coefficent of 4 means that the available steps to reach the target are "initialDistance (between agent and target) * 4" (for more info check LTApproach)
//Reusing TSHints from LTApproach with some additions
TSHints.Add(MOVING_VELOCITY, 1);
TSHints.Add(ELLIPSE_SIZE, 0.55f);
TSHints.Add(STEPS_TAKEN_FOR_ONE_CIRCLE, 600);
TSHints.Add(AVOIDING_AGENT, 0);
TSProgression.Clear(); // Clearing TSProgression that was declared in LTApproach before filling custom one
TSProgression.Add(TSHints.Clone());
TSProgression.Add(
new TrainingSetHints {
{ ELLIPSE_SIZE, 0.55f },
{ STEPS_TAKEN_FOR_ONE_CIRCLE, 500 }
});
TSProgression.Add(
new TrainingSetHints {
{ ELLIPSE_SIZE, 0.6f },
{ STEPS_TAKEN_FOR_ONE_CIRCLE, 400 }
});
TSProgression.Add(
new TrainingSetHints {
{ ELLIPSE_SIZE, 0.7f },
{ STEPS_TAKEN_FOR_ONE_CIRCLE, 300 }
});
TSProgression.Add(
new TrainingSetHints {
{ ELLIPSE_SIZE, 0.7f },
{ STEPS_TAKEN_FOR_ONE_CIRCLE, 200 }
});
TSProgression.Add(
new TrainingSetHints {
{ ELLIPSE_SIZE, 0.8f },
{ STEPS_TAKEN_FOR_ONE_CIRCLE, 100 }
});
TSProgression.Add(
new TrainingSetHints {
{ MOVING_VELOCITY, 2 },
{ ELLIPSE_SIZE, 0.65f },
{ STEPS_TAKEN_FOR_ONE_CIRCLE, 1000 },
{ AVOIDING_AGENT, 1 },
{ DISTANCE_BONUS_COEFFICENT, 7f },
{ TSHintAttributes.IMAGE_TEXTURE_BACKGROUND, 1f }
});
}
public LTShapeGroups(SchoolWorld w)
: base(w)
{
// TODO Add TSHints
TSProgression.Add(TSHints.Clone());
// TODO progression
SetHints(TSHints);
}
public virtual TrainingResult EvaluateStep()
{
// Check for unit completion
bool wasUnitSuccessful = true;
bool inProgress = (SchoolWorld.IsEmulatingUnitCompletion() && !SchoolWorld.EmulateIsTrainingUnitCompleted(out wasUnitSuccessful)) ||
(!SchoolWorld.IsEmulatingUnitCompletion() && !DidTrainingUnitComplete(ref wasUnitSuccessful));
if (inProgress)
{
// The unit is still in progress
return(TrainingResult.TUInProgress);
}
// otherwise the unit is over
CurrentNumberOfAttempts++;
// Check for task failure
if (TSHints.ContainsKey(TSHintAttributes.MAX_NUMBER_OF_ATTEMPTS) &&
CurrentNumberOfAttempts >= TSHints[TSHintAttributes.MAX_NUMBER_OF_ATTEMPTS])
{
// Too many attempts
Reward = -1.0f;
return(TrainingResult.FailedLT);
}
if (wasUnitSuccessful)
{
Reward = 1.0f;
CurrentNumberOfSuccesses++;
}
else
{
Reward = -1.0f;
CurrentNumberOfSuccesses = 0;
}
MyLog.Writer.WriteLine(
MyLogLevel.INFO,
GetTypeName() +
" unit ends with result: " +
(wasUnitSuccessful ? "success" : "fail") +
". " +
CurrentNumberOfSuccesses + " successful attempts in row, " +
NumberOfSuccessesRequired + " required.");
// Check for level completion
if (CurrentNumberOfSuccesses < NumberOfSuccessesRequired)
{
// The level is still in progress
return(TrainingResult.FinishedTU);
}
return(TrainingResult.FinishedLevel);
}
public virtual void Init()
{
CurrentNumberOfAttempts = 0;
CurrentNumberOfSuccesses = 0;
CurrentLevel = 0;
TSHints.Set(TSProgression[CurrentLevel]);
SetHints(TSHints);
IsInitialized = true;
}
public LTMultipleTargetsSequence(SchoolWorld w)
: base(w)
{
TSHints[TSHintAttributes.DEGREES_OF_FREEDOM] = 2; // Set degrees of freedom to 2: move in 4 directions (1 means move only right-left)
TSHints[TSHintAttributes.NUMBER_OF_DIFFERENT_OBJECTS] = 1;
TSHints[TSHintAttributes.MAX_NUMBER_OF_ATTEMPTS] = 10000;
TSHints.Add(SEQUENCE_LENGTH, 1);
TSHints.Add(TIMESTEPS_LIMIT, 500); // Training unit fails if the TIMESTEP_LIMIT is reached, this is used to avoid that the agent stands without moving
TSHints.Add(DISTANCE_BONUS_COEFFICENT, 3.0f); // Manhattan distance is calculated to estimate how many steps are needed to finish the sequence, DISTANCE_BONUS_COEFFICENT can be used to increase that number, if for example it's 2.0f then TIMESTEPS_LIMIT is (2 * EstimatedManahattanDistance)
TSHints.Add(TSHintAttributes.IMAGE_TEXTURE_BACKGROUND, 1f);
TSProgression.Add(TSHints.Clone());
TSProgression.Add(
new TrainingSetHints {
{ SEQUENCE_LENGTH, 2 },
{ DISTANCE_BONUS_COEFFICENT, 3f }
});
TSProgression.Add(
new TrainingSetHints {
{ SEQUENCE_LENGTH, 3 },
{ DISTANCE_BONUS_COEFFICENT, 3f }
});
TSProgression.Add(
new TrainingSetHints {
{ SEQUENCE_LENGTH, 4 },
{ DISTANCE_BONUS_COEFFICENT, 2.5f }
});
TSProgression.Add(
new TrainingSetHints {
{ SEQUENCE_LENGTH, 5 },
{ DISTANCE_BONUS_COEFFICENT, 2.0f }
});
TSProgression.Add(
new TrainingSetHints {
{ DISTANCE_BONUS_COEFFICENT, 1.5f }
});
TSProgression.Add(
new TrainingSetHints {
{ DISTANCE_BONUS_COEFFICENT, 1.0f },
});
}
public LTObstaclesTargetInSight(SchoolWorld w)
: base(w)
{
TSHints.Clear();
TSProgression.Clear();
TSHints.Add(TIMESTEPS_LIMIT, 200);
TSHints.Add(TSHintAttributes.MAX_NUMBER_OF_ATTEMPTS, 10000);
TSHints.Add(TSHintAttributes.IMAGE_NOISE, 0);
TSHints.Add(TSHintAttributes.IMAGE_TEXTURE_BACKGROUND, 0);
TSProgression.Add(TSHints.Clone());
TSProgression.Add(TIMESTEPS_LIMIT, 100);
TSProgression.Add(TSHintAttributes.IMAGE_NOISE, 1);
TSProgression.Add(TIMESTEPS_LIMIT, 50);
TSProgression.Add(TSHintAttributes.IMAGE_TEXTURE_BACKGROUND, 1);
}
public LTObstacles(SchoolWorld w)
: base(w)
{
TSHints[TSHintAttributes.IMAGE_NOISE] = 0;
TSHints[TSHintAttributes.DEGREES_OF_FREEDOM] = 2; // Set degrees of freedom to 2: move in 4 directions (1 means move only right-left)
TSHints[TSHintAttributes.NUMBER_OF_DIFFERENT_OBJECTS] = 1;
TSHints[TSHintAttributes.MAX_NUMBER_OF_ATTEMPTS] = 10000;
TSHints.Add(OBSTACLES_LEVEL, 1);
TSHints.Add(TIMESTEPS_LIMIT, 800);
TSProgression.Add(TSHints.Clone());
TSProgression.Add(OBSTACLES_LEVEL, 2);
TSProgression.Add(OBSTACLES_LEVEL, 3);
TSProgression.Add(OBSTACLES_LEVEL, 4);
TSProgression.Add(OBSTACLES_LEVEL, 5);
TSProgression.Add(OBSTACLES_LEVEL, 6);
}
