def decision_function(A_SUCCESS, A_FAIL, B_SUCCESS, B_FAIL):
decision_dict = {'DECISION': 'continue', 'ESTIMATED_DIFFERENCE': None}
N_A = A_SUCCESS + A_FAIL
N_B = B_SUCCESS + B_FAIL
N_BOTH = N_A + N_B
if N_BOTH % JUMP_ == 0 or N_BOTH >= MAX_TEST_SIZE:
A_ALPHA_POST = ALPHA + A_SUCCESS
A_BETA_POST = BETA + A_FAIL
B_ALPHA_POST = B_PRIORS[0] + B_SUCCESS
B_BETA_POST = B_PRIORS[1] + B_FAIL
POSTERIOR = sample_posterior(A_ALPHA_POST, A_BETA_POST,
B_ALPHA_POST, B_BETA_POST, N_A, N_B,
HORIZON_LENGTH - N_BOTH, N_SAMPLES)
LOSS_DIFF = POSTERIOR['LOSS_B'] - POSTERIOR['LOSS_A']
if abs(LOSS_DIFF) > THRESHOLD or N_BOTH >= MAX_TEST_SIZE:
decision_dict['ESTIMATED_DIFFERENCE'] = POSTERIOR['P_DIFF']
if LOSS_DIFF > 0: # Loss for B is greater than loss for A
decision_dict['DECISION'] = 'A'
else:
decision_dict['DECISION'] = 'B'
return decision_dict
def decision_function(A_SUCCESS, A_FAIL, B_SUCCESS, B_FAIL):
decision_dict = {
'DECISION': 'continue',
'ESTIMATED_DIFFERENCE': None
}
N_A = A_SUCCESS + A_FAIL
N_B = B_SUCCESS + B_FAIL
N_BOTH = N_A + N_B
if N_BOTH % JUMP == 0 or N_BOTH > MAX_TEST_SIZE:
A_ALPHA_POST = ALPHA + A_SUCCESS
A_BETA_POST = BETA + A_FAIL
B_ALPHA_POST = B_PRIORS[0] + B_SUCCESS
B_BETA_POST = B_PRIORS[1] + B_FAIL
POSTERIOR = sample_posterior(A_ALPHA_POST, A_BETA_POST,
B_ALPHA_POST, B_BETA_POST, N_A,
N_B, HORIZON_LENGTH - N_BOTH,
N_SAMPLES)
if HARD:
if POSTERIOR['P_WIN'] > P_THRESHOLD_:
decision_dict['DECISION'] = 'B'
decision_dict['ESTIMATED_DIFFERENCE'] = POSTERIOR[
'P_DIFF']
elif POSTERIOR['P_WIN'] < (
1 - P_THRESHOLD_) or N_BOTH >= MAX_TEST_SIZE:
decision_dict['DECISION'] = 'A'
decision_dict['ESTIMATED_DIFFERENCE'] = POSTERIOR[
'P_DIFF']
else:
if POSTERIOR['P_WIN'] > P_THRESHOLD_ or POSTERIOR[
'P_WIN'] < (
1 - P_THRESHOLD_) or N_BOTH >= MAX_TEST_SIZE:
decision_dict['ESTIMATED_DIFFERENCE'] = POSTERIOR[
'P_DIFF']
if POSTERIOR[
'P_WIN'] > 0.5: # Loss for B is greater than loss for A
decision_dict['DECISION'] = 'B'
else:
decision_dict['DECISION'] = 'A'
return decision_dict