if G_DIFF == False and D_DIFF == True:
sys.exit()
tf.set_random_seed(SEED)
np.random.seed(SEED)
##############################################################################
# prepare data
FILE_NAME = 'pickled_data_ppgan_{}'.format(DATA)
if not os.path.isfile(FILE_NAME):
if DATA == 'gaussian' and False: #QQ_plot for gaussian is not good as hawkes,selfcorrecting, perhaps that simulating is not good.
intensityGaussian = IntensitySumGaussianKernel(3, [3, 7, 11],
[1, 1, 1], [2, 3, 2])
real_sequences = generate_sample(intensityGaussian, T, 20000)
sequence2file(real_sequences, 'gaussian')
else:
real_sequences = file2sequence(DATA)
lambda0 = np.mean([len(item) for item in real_sequences]) / T
intensityPoisson = IntensityHomogenuosPoisson(lambda0)
fake_sequences = generate_sample(intensityPoisson, T, 20000)
pickle.dump([real_sequences, fake_sequences], open(FILE_NAME, 'wb'))
else:
real_sequences, fake_sequences = pickle.load(open(FILE_NAME, 'rb'))
print(
np.mean([len(item) for item in real_sequences]) / T,
np.mean([len(item) for item in fake_sequences]) / T)
if not REAL_DATA:
real_sequences = real_sequences[:SEQ_NUM]
else:
REAL_DATA = False
tf.set_random_seed(SEED)
np.random.seed(SEED)
##############################################################################
# prepare data
FILE_NAME = 'pickled_data_{}'.format(DATA)
if not os.path.isfile(FILE_NAME):
if DATA == 'gaussian': #QQ_plot for gaussian is not good as hawkes,selfcorrecting, perhaps that simulating is not good.
intensityGaussian = IntensitySumGaussianKernel(3, [3, 7, 11],
[1, 1, 1], [2, 3, 2])
real_sequences = generate_sample(intensityGaussian, T, 20000)
sequence2file(real_sequences, 'gaussian')
else:
real_sequences = file2sequence(DATA)
#lambda0 = np.mean([len(item) for item in real_sequences])/T
#intensityPoisson = IntensityHomogenuosPoisson(lambda0)
#fake_sequences = generate_sample(intensityPoisson, T, 20000)
pickle.dump(real_sequences, open(FILE_NAME, 'wb'))
else:
real_sequences = pickle.load(open(FILE_NAME, 'rb'))
print((np.mean([len(item) for item in real_sequences]) / T))
if not REAL_DATA:
real_sequences = real_sequences[:SEQ_NUM]
real_iterator = PaddedDataIterator(real_sequences, T, MARK, D_DIFF)
plt.close()
if not REAL_DATA and DATA!="rmtpp":
integral_intensity = get_integral(sequences_generator, DATA)
integral_intensity = np.asarray(integral_intensity)
fig = sm.qqplot(integral_intensity, stats.expon, distargs=(), loc=0, scale=1,line='45')
res,slope_intercept = stats.probplot(integral_intensity, dist=stats.expon)
plt.grid()
fig.savefig('out/{}/{}.png'.format(saved_file,it))
plt.close()
if np.abs(1-slope_intercept[0])<1e-1 and deviation<1e-1:
stop_indicator = True
elif deviation<1e-2:
stop_indicator = True
if it == ITERS-1 or stop_indicator:
sequences_generator = []
for _ in range(int(20000/BATCH_SIZE)):
sequences_gen = sess.run(fake_data,feed_dict={Z:fake_batch[0], fake_seqlen:fake_batch[1]})
shape_gen = sequences_gen.shape
sequences_gen = np.reshape(sequences_gen,(shape_gen[0],shape_gen[1]))
if D_DIFF:
sequences_gen = np.cumsum(sequences_gen,axis=1)
sequences_gen = sequence_filter(sequences_gen,fake_batch[1]) # remove padding tokens
sequences_generator +=sequences_gen
sequence2file(sequences_generator, 'wgan_{}_{}_{}_{}'.format(DATA,SEQ_NUM,ITERATION,LAMBDA_LP))
break
else:
REAL_DATA = False
tf.set_random_seed(SEED)
np.random.seed(SEED)
##############################################################################
# prepare data
FILE_NAME = 'pickled_data_ppgan_{}'.format(DATA)
if not os.path.isfile(FILE_NAME):
if DATA == 'gaussian': #QQ_plot for gaussian is not good as hawkes,selfcorrecting, perhaps that simulating is not good.
intensityGaussian = IntensitySumGaussianKernel(3, [3, 7, 11],
[1, 1, 1], [2, 3, 2])
real_sequences = generate_sample(intensityGaussian, T, 20000)
sequence2file(real_sequences, 'gaussian')
else:
real_sequences = file2sequence(DATA)
lambda0 = np.mean([len(item) for item in real_sequences]) / T
intensityPoisson = IntensityHomogenuosPoisson(lambda0)
fake_sequences = generate_sample(intensityPoisson, T, 2000)
pickle.dump([real_sequences, fake_sequences], open(FILE_NAME, 'wb'))
else:
real_sequences, fake_sequences = pickle.load(open(FILE_NAME, 'rb'))
real_sequences, _ = pickle.load(open(FILE_NAME, 'rb'))
print((np.mean([len(item) for item in real_sequences]) / T),
((np.mean([len(item) for item in fake_sequences]) / T)))
if not REAL_DATA:
real_sequences = real_sequences[:SEQ_NUM]