def fit_ARIMA():
init_bic = float("inf")
init_p = 0
init_q = 0
init_d = 1
maxLag = 8
with open('dev_data.json', 'r') as f:
raw_data = json.load(f)
data = []
for index in range(len(raw_data)):
if index % 10 == 0:
data.extend([float(di[1]) for di in raw_data[index]['data']])
# data.extend([float(di[1]) for di in d['data']])
init_properModel = None
for p in np.arange(0, maxLag):
for q in np.arange(0, maxLag):
for d in np.arange(1, 3):
print('test p = %d, d = %d, q = %d' % (p, d, q))
model = ARIMA(data, order=(p, d, q))
## bug cuased by statsmodels lib, if do not set them then error occurs during saving
model.dates = None
model.freq = None
model.missing = None
try:
results_ARIMA = model.fit(disp=-1, method='css')
except:
continue
bic = results_ARIMA.bic
print('current bic = %.4f' % (bic))
if bic < init_bic:
print(
'It BEATS the previous best p, d and q, current p = %d, d = %d, q = %d'
% (p, d, q))
init_p = p
init_q = q
init_d = d
init_properModel = results_ARIMA
init_bic = bic
results_ARIMA.save('ARIMA_best')
print('Saved the best model')
print('p is :' + str(init_p))
print('q is :' + str(init_q))
print('d is :' + str(init_d))
len(lines))
le = LabelEncoder()
le.fit(list(alphabet))
print("Fitted alphabet")
seq = []
max_np_len = 200000
for i in range(0, len(words), max_np_len):
print(i)
seq.extend(list(le.transform(words[i:i + max_np_len])))
for i in range(len(seq)):
seq[i] = float(seq[i])
model = ARIMA(seq, order=(2, 1, 0))
model_fit = model.fit(disp=0)
output = model_fit.forecast()
prediction = le.inverse_transform(
[int(symbol.flatten()[0]) for symbol in output])
print("Single forecast:", prediction)
print("Fit features")
model.dates, model.freq, model.missing = None, None, None
with open(args.o + "arima.model_fit", "wb") as f:
pickle.dump(model_fit, f)
with open(args.o + "arima.model", "wb") as f:
pickle.dump(model, f)
warn("Output written to arima.model and arima.model_fit")
print(datetime.datetime.now())
