class EmpiricalLearningProgress():
def __init__(self, task_size):
self.interest_knn = BufferedDataset(1,
task_size,
buffer_size=2000,
lateness=0)
#self.window_size = 1000
def get_lp(self, task, competence):
interest = 0
if len(self.interest_knn) > 5:
# compute learning progre ss for new task
dist, idx = self.interest_knn.nn_y(task)
# closest_previous_task = previous_tasks[idx]
closest_previous_task = self.interest_knn.get_y(idx[0])
closest_previous_task_competence = self.interest_knn.get_x(idx[0])
# print 'closest previous task is index:%s, val: %s' % (idx[0], closest_previous_task)
# compute Progress as absolute difference in competence
progress = closest_previous_task_competence - competence
interest = np.abs(progress)
# add to database
self.interest_knn.add_xy(competence, task)
return interest
class EmpiricalALPComputer():
def __init__(self, task_size, max_size=None, buffer_size=500):
self.alp_knn = BufferedDataset(1,
task_size,
buffer_size=buffer_size,
lateness=0,
max_size=max_size)
def compute_alp(self, task, reward):
alp = 0
if len(self.alp_knn) > 5:
# Compute absolute learning progress for new task
# 1 - Retrieve closest previous task
dist, idx = self.alp_knn.nn_y(task)
# 2 - Retrieve corresponding reward
closest_previous_task_reward = self.alp_knn.get_x(idx[0])
# 3 - Compute alp as absolute difference in reward
lp = reward - closest_previous_task_reward
alp = np.abs(lp)
# Add to database
self.alp_knn.add_xy(reward, task)
return alp
def __init__(self, task_size):
self.interest_knn = BufferedDataset(1,
task_size,
buffer_size=2000,
lateness=0)
def __init__(self, task_size, max_size=None, buffer_size=500):
self.alp_knn = BufferedDataset(1, task_size, buffer_size=buffer_size, lateness=0, max_size=max_size)