DAgger Aggregation Step

Problem

Aggregate a new DAgger round into the training dataset.

Append the newly queried states and expert actions to the old data and report the fraction of the merged dataset that comes from the new expert labels. Return [merged_states, merged_actions, expert_fraction]; an empty merge gives fraction 0.0.

Hint

DAgger grows the dataset with expert labels on the learner's own visited states.

Test cases

case 1

in: [["s0"],[0],["s1","s2"],[1,0]]

expected: [["s0","s1","s2"],[0,1,0],0.6666666666666666]

case 2

in: [[],[],[],[]]

expected: [[],[],0]

case 3

in: [[],[],["x"],[2]]

expected: [["x"],[2],1]

Starter codePython

def dagger_aggregation_step(states, actions, new_states, new_actions):
    # Your code here
    pass

3 test cases · run your code against them in the editor