def test_non_seasonal_arima3(self):
ts_data = self.getData4()
f_name='non_seasonal_arima3.pmml'
model = ARIMA(ts_data,order=(1, 0, 1))
result = model.fit(trend = 'c', method = 'css-mle')
StatsmodelsToPmml(result, f_name, description="A test model")
self.assertEqual(os.path.isfile(f_name),True)
def test_non_seasonal_arima1(self):
ts_data = self.getData4()
f_name='non_seasonal_arima1.pmml'
model = ARIMA(ts_data,order=(9, 2, 0))
result = model.fit(trend = 'c', method = 'css-mle')
StatsmodelsToPmml(result, f_name, model_name="arima_920")
self.assertEqual(os.path.isfile(f_name),True)
def test_non_seasonal_arima8(self):
ts_data = self.statsmodels_data_helper.get_non_seasonal_data()
f_name = 'non_seasonal_arima8.pmml'
model = ARIMA(ts_data, order=(5, 1, 2))
result = model.fit(trend='c', method='mle')
ArimaToPMML(result, f_name, conf_int=[80, 95])
self.assertEqual(self.schema.is_valid(f_name), True)
def test_non_seasonal_arima2(self):
ts_data = self.getData4()
f_name = 'non_seasonal_arima2.pmml'
model = ARIMA(ts_data, order=(9, 2, 3))
result = model.fit(trend='nc', method='css-mle')
ArimaToPMML(result, f_name, description="A test model")
self.assertEqual(os.path.isfile(f_name), True)
def test_non_seasonal_arima1(self):
ts_data = self.statsmodels_data_helper.get_non_seasonal_data()
f_name = 'non_seasonal_arima1.pmml'
model = ARIMA(ts_data, order=(9, 2, 0))
result = model.fit(trend='c', method='css-mle')
ArimaToPMML(result, f_name)
self.assertEqual(self.schema.is_valid(f_name), True)
def test_non_seasonal_arima7(self):
ts_data = self.statsmodels_data_helper.get_non_seasonal_data()
f_name = 'non_seasonal_arima7.pmml'
model = ARIMA(ts_data, order=(5, 1, 2))
result = model.fit(trend='nc', method='mle')
StatsmodelsToPmml(result, f_name)
self.assertEqual(self.schema.is_valid(f_name), True)
def predict_by_samples(data,
test_data,
train_length,
num_sample,
output_steps,
lag_order=1,
if_sample=True):
warnings.filterwarnings("ignore")
test_num, data_dim = test_data.shape
if not if_sample:
num_sample = data_dim
#widgets = ['Train: ', Percentage(), ' ', Bar('-'), ' ', ETA()]
#pbar = ProgressBar(widgets=widgets, maxval=test_data.shape[0]-output_steps).start()
error_all = []
index_all = np.zeros([test_num - output_steps, num_sample], dtype=np.int32)
valid_num = np.zeros(test_num - output_steps, dtype=np.int32)
real = np.zeros([test_num - output_steps, num_sample])
predict = np.zeros([test_num - output_steps, num_sample])
#
samples = np.arange(data_dim)
#
for t in range(test_num - output_steps):
#pbar.update(t)
if t % 100 == 0:
print(t)
error_index = []
if if_sample:
samples = np.random.randint(data_dim, size=num_sample)
for r in range(num_sample):
# t: which time slot
# i: which station
i = samples[r]
train_df = data[t:train_length + t, i]
try:
results = ARIMA(train_df,
order=(lag_order, 0, 1)).fit(trend='nc',
disp=0)
except:
error_index.append(r)
continue
#pre, _, _ = results.forecast(output_steps)
#pre = results.predict(train_length, train_length+output_steps)
pre = results.predict(train_length, train_length)
test_real = test_data[t:t + output_steps, i]
real[t, r] = test_real
#print(pre)
predict[t, r] = pre
index_all[t] = samples
error_all.append(error_index)
valid_num[t] = num_sample - len(error_index)
#pbar.finish()
return real, predict, index_all, error_all, valid_num
def train(self, array_X, array_Y):
self.train_X = array_X
self.train_Y = array_Y
array = numpy.concatenate((numpy.array([array_Y]).T, array_X), axis=1)
idx = pd.date_range('20130101', periods=48000)
#xxx= pd.DataFrame(data=numpy.array([array_Y]),index=idx)
xxx = pd.DataFrame(data=array, index=idx)
model = ARIMA(endog=xxx, order=(0, 1, 1))
fit = model.fit()
res = fit.fittedvalues.values[:, 0]
res = numpy.hstack((res[0], res))
return res
def arima(time_serie, Npt, order, simple_ticker, fit: bool, save: bool, load: bool, path):
filename = path+'/Resources/TunnedModels/' + simple_ticker + '_arima_2020_07_31.pickle'
if fit:
arima_fit = ARIMA(time_serie.values.astype('float'), order).fit(disp=False)
if save:
# save the model to disk
pickle.dump(arima_fit, open(filename, 'wb'))
elif load:
# load the model from disk
arima_fit = pickle.load(open(filename, 'rb'))
else:
print("Error: fit and load parameters are both false.")
sys.exit(-1)
forecast_arima = [x for x in arima_fit.forecast(Npt)[0]] # [0]= predições
return pd.Series(forecast_arima)
def predict(self, test_X, test_Y):
predictions = numpy.empty(0)
array_train = numpy.concatenate(
(numpy.array([self.train_Y]).T, self.train_X), axis=1)
array_test = numpy.concatenate((numpy.array([test_Y]).T, test_X),
axis=1)
for t in range(0, test_Y.shape[0]):
array = numpy.vstack((array_train, array_test[:t]))
model = ARIMA(endog=pd.DataFrame(data=array))
fit = model.fit()
lag = fit.k_ar
pred = fit.forecast(array[-lag:], 1)[0]
predictions = numpy.append(predictions, pred[0])
return predictions
def build_model(self):
"""Build ARIMA model with 3 hyperparameters
See https://en.wikipedia.org/wiki/Autoregressive_integrated_moving_average
for detailed information
"""
self.p = self.para['p']
self.d = self.para['d']
self.q = self.para['q']
self.model = ARIMA(self.train,
order=(self.p, self.d, self.q),
exog=self.exog_train)
