shivindass / inverse_rl Goto Github PK

for i in range(25):
    optimalAction = policy[i]
    tempTransProbMatrix= gamma * trans_probs[:, :, optimalAction]
    test = np.identity(25)-tempTransProbMatrix
    tempInverse = np.linalg.inv(np.identity(25) - tempTransProbMatrix)
    print(tempInverse.shape)
    
    for j in range(4):
        if(j != optimalAction):
            test_c = trans_probs[i, :, optimalAction]
            print(test_c.shape)

            condition= - np.dot(trans_probs[i, :, optimalAction] - trans_probs[i, :, j], tempInverse)
            print(condition.shape)
            conditions.append(condition)
print(len(conditions), len(condition[0]))
equality=np.zeros(625)
for i in range(25):
    c[25:2 * 25] = -1
    c[2 * 25:] = l1
print(c.shape)
print(c)
conditions=np.array(conditions)
print(conditions)
conditions=np.reshape(conditions,[625,75])
print(conditions)
print(len(c),conditions.shape)
rewards=linprog(c, A_ub=conditions, b_ub=equality)

#rewards = rewards/max(rewards)
return rewards`