def priceActual():
for k,v in priceDic.items():
print(k,v)
series = pd.read_csv('../Data/Final/Wholesale/'+k,names = [0.1],index_col=0,header=None)
near_file = v[:-4]+'.npy'
near = np.load('../Data/Results/Near/'+near_file)
daysToForecast = 30
series_train = np.squeeze(series.values)
n = len(series_train)
near_train = near[:n]
near_test = near[-30:]
trans = FourierFeaturizer(365.25, 1)
y_prime, exogen = trans.fit_transform(series_train)
exogen = exogen.mul(pd.Series(series_train),axis=0)
futureExog = trans.transform(y = series_train, n_periods = 30)
futureExog = pd.DataFrame(futureExog[1])
futureExog = futureExog.mul(pd.Series(near_test),axis=0)
exogen['near'] = near_train
futureExog['near'] = near_test
#print('MODel searching')
model=pm.arima.auto_arima(series_train, exogenous = exogen, start_p=0, d=None, start_q=0, max_p=3, max_d=1, max_q=3,start_P=0, D=None, start_Q=0, max_P=2, max_D=1, max_Q=2,suppress_warnings =True,seasonal=True,max_order=4,m=7,stepwise=True)
model.fit(series_train)
pred = (model.predict(daysToForecast,exogenous = futureExog))
series = np.concatenate((series_train,pred),axis=0)
series = pd.DataFrame(series)
series.index = pd.date_range(start = '2006-01-01',periods = len(series))
fileName = '../Data/Results/Actual/'+str(k)
#print(fileName)
series.to_csv(fileName)
def forecast(series,near):
train_near = np.squeeze(near.values[:-30])
train_series = np.squeeze(series.values[:len(train_near)])
test_near = np.squeeze(near.values[-30:])
test_series = np.squeeze(series.values[-30:])
df = pd.DataFrame(columns=['aic','nonSeasonal','seasonal','k'])
for i in range(25):
print('value of i is:',i)
nonSeasonalParams = ch(nonSeasonal)
seasonalParams = ch(seasonal)
val = sum(list(nonSeasonalParams)) + sum(list(seasonalParams))
#try:
if(val>8):
continue
seasonalParams = seasonalParams + (7,)
#print(nonSeasonalParams,seasonalParams)
try:
trans = FourierFeaturizer(365.25, 1)
y_prime, exogen = trans.fit_transform(train_series)
#exogen = exogen.mul(pd.Series(train_series),axis=0)
exogen['near'] = train_near
model = sm.tsa.statespace.SARIMAX(endog = train_series, exog = train_near, order = nonSeasonalParams, seasonal_order = seasonalParams,initialization='approximate_diffuse',enforce_stationarity=False)
res = model.fit(disp=False)
#print(res.aic)
to_append = [res.aic,nonSeasonalParams,seasonalParams,1]
a_series = pd.Series(to_append, index = df.columns)
df = df.append(a_series, ignore_index=True)
except:
print('inside except block now.....')
x = pd.Series([10000000,(1,1,1),(1,1,1),1],index = df.columns)
df = df.append(x, ignore_index=True)
x = pd.Series([10000000,(1,1,1),(1,1,1),1],index = df.columns)
df = df.append(x, ignore_index=True)
print('df is :',df)
dx = (df[df.aic == df.aic.min()])
dx.reset_index(inplace=True,drop=True)
print('baest parameters are:',dx)
value = [dx.iloc[0][0],dx.iloc[0][1],dx.iloc[0][2],dx.iloc[0][3]]
print('final parameters are:',value)
return value
header=None)
near = pd.read_csv('../Data/Final/Wholesale/YeolaPrice.csv',
names=[0.1],
index_col=0,
header=None)
train_near = np.squeeze(near.values[:-30])
train_series = np.squeeze(series.values[:len(train_near)])
test_near = np.squeeze(near.values[-30:])
test_series = np.squeeze(series.values[-30:])
for k in range(1, 4):
print(k)
trans = FourierFeaturizer(365.25, k)
y_prime, exogen = trans.fit_transform(train_series)
exogen = exogen.mul(pd.Series(train_series), axis=0)
exogen['near'] = train_near
model = pm.arima.auto_arima(train_series,
exogenous=pd.DataFrame(exogen),
start_p=0,
d=None,
start_q=0,
max_p=5,
max_d=2,
max_q=5,
start_P=0,
D=None,
start_Q=0,
max_P=5,
max_D=2,