reward_types = [True]
res_r={}
for reward_type in reward_types:
res_b={}
for beta in betas:
thr=False
res_e={}
res_e[0]={"cnvg_val": trainer.test_curr_discriminator_batch(), "steps": 0}
for epsilon in eps:
print("t: "+str(n_authorized)+", snr: "+str(snr)+", eps: "+str(epsilon)+", beta: "+str(beta)+", reward: "+str(reward_type))
if thr:
res_e[epsilon]={"cnvg_val": trainer.test_curr_discriminator_batch(), "steps": 0}
else:
trainer.reset_generator(lr=1e-3, epsilon=epsilon, beta=beta, binary_reward=reward_type)
trainer.reflect_pre_train()
cnvg_val, steps = trainer.train_loop(steps=1000, sensitivity=sensitivity)
res_e[epsilon]={"cnvg_val": cnvg_val, "steps": steps}
if(cnvg_val < 0.01):
thr=True
res_b[beta]=res_e
res_r[reward_type]=res_b
res_t[snr]=res_r
data_pickle[n_authorized]=res_t
if(dump_pickle_each_loop): pickle.dump(data_pickle, open(pickle_path, "wb" ))
res_b[beta]=res_e
res_r[reward_type]=res_b
res_t[snr]=res_r
data_pickle[n_authorized]=res_t
if(dump_pickle_each_loop): pickle.dump(data_pickle, open(pickle_path, "wb" ))
sig_im_train_re_t = sig_im_train.reshape((-1, 2))
sig_im_valid_re = sig_im_valid.reshape((-1, 1, 2))
sig_im_valid_re_t = sig_im_valid.reshape((-1, 2))
sig_im_test_re = sig_im_test.reshape((-1, 1, 2))
trainer.pre_train_generator(
g_epochs=100,
g_pre_data_train=[sig_im_train_re, sig_im_train_re_t],
g_pre_data_valid=[sig_im_valid_re, sig_im_valid_re_t],
lr=1e-3)
pred_curr = trainer.predict_curr_discriminator()
assert (pred_curr <= 0.4)
[0., 0.02, 0.05, 0.07, 0.08, 0.10, 0.12, 0.2]
eps = [0., 0.02, 0.05, 0.06, 0.07, 0.08, 0.09, 0.10, 0.12, 0.15, 0.2]
betas = [0, 1000, 5000, 7500, 10000, 12500, 15000, 20000]
reward_types = [True, False]
for epsilon in eps:
for beta in betas:
for reward_type in reward_types:
trainer.reset_generator(lr=1e-3,
epsilon=0.05,
beta=5000,
binary_reward=True)
trainer.reflect_pre_train()
trainer.train(steps=1000)