epsilon=epsilon)
opt_action_sarsa = []
for i in range(sarsa_Q_star_est.shape[0]):
opt_action_sarsa.append(np.argmax(sarsa_Q_star_est[i, :]))
df_final = trajs_train[[
'state_done', 'cumulative_overdue_early_difference', 'gender', 'amount',
'num_loan', 'duration', 'year_ratio', 'diff_city', 'marriage', 'kids',
'month_in', 'housing', 'edu', 'motivation'
]].drop_duplicates()
df_final['opt_action_sarsa'] = opt_action_sarsa
qlearning_star_est, qLearning_reward = qlearning(trajectory,
nS,
nA,
alpha=0.4,
gamma=gamma,
epsilon=epsilon)
opt_action_qlearning = []
for i in range(qlearning_star_est.shape[0]):
opt_action_qlearning.append(np.argmax(qlearning_star_est[i, :]))
df_final['opt_action_qlearning'] = opt_action_qlearning
trajs_test = trajs_test.merge(
df_final,
on=[
'state_done', 'cumulative_overdue_early_difference', 'gender',
'amount', 'num_loan', 'duration', 'year_ratio', 'diff_city',
'marriage', 'kids', 'month_in', 'housing', 'edu', 'motivation'
],
behavior_policy2,
nS=nS2,
nA=nA2,
n=1,
alpha=0.005)
print(
'###########################################################################'
)
print(sarsa_Q_star_est2[-10:, -10:])
# print(sarsa_reward2[-50:])
# print('estimated policy is')
# print(sarsa_pi_star_est2)
qlearning1, qLearning1_reward = qlearning(trajs1,
nS1,
nA1,
alpha=0.4,
gamma=0.999,
epsilon=0.9)
print(
'###########################################################################'
)
print(qlearning1[-10:, -10:])
# print(qLearning1_reward[-50:])
qlearning2, qlearning2_reward = qlearning(trajs2,
nS2,
nA2,
alpha=0.4,
gamma=0.999,
epsilon=0.9)
print(
'###########################################################################'
#for i in range(sarsa_Q_star_est1.shape[0]):
# opt_action1_sarsa.append(np.argmax(sarsa_Q_star_est1[i,:]))
df1_final = trajs1_pd_train[[
'state_done', 'cumulative_overdue_early_difference', 'gender', 'amount',
'num_loan', 'duration', 'year_ratio', 'diff_city', 'marriage', 'kids',
'month_in', 'housing', 'edu', 'motivation'
]].drop_duplicates()
#df1_final['opt_action_sarsa'] = opt_action1_sarsa
print('type of trajectory is ', type(trajs1))
recommended_actions = []
qlearning1, qLearning1_reward = qlearning(trajs1,
nS1,
nA1,
recommended_Q=[],
alpha=0.4,
gamma=gamma,
epsilon=0.9)
opt_action1_qlearning = []
for i in range(qlearning1.shape[0]):
opt_action1_qlearning.append(np.argmax(qlearning1[i, :]))
df1_final['opt_action_qlearning'] = opt_action1_qlearning
trajs1_pd_test = trajs1_pd_test.merge(
df1_final,
on=[
'state_done', 'cumulative_overdue_early_difference', 'gender',
'amount', 'num_loan', 'duration', 'year_ratio', 'diff_city',
'marriage', 'kids', 'month_in', 'housing', 'edu', 'motivation'