def ANN(error_train, error_test, data, split_len):
train_data = data.iloc[:split_len, :]
test_data = data.iloc[split_len:, :]
x_train = error_train[1:]
y_train = train_data[1:]
x_test = error_test[1:]
y_test = test_data[1:]
model_ann = Sequential()
model_ann.add(Dense(4, input_dim=1, activation="sigmoid"))
model_ann.add(Dense(4))
model_ann.add(Dense(1))
model_ann.compile(loss="mean_squared_error", optimizer="adam")
early_stop = EarlyStopping(monitor="loss", patience=25, verbose=1)
model_ann.fit(x_train,
y_train,
epochs=200,
callbacks=[early_stop],
batch_size=10,
verbose=1,
shuffle=False)
y_test_pred = model_ann.predict(x_test)
y_train_pred = model_ann.predict(x_train)
print("The MSE score on the Train set is:\t{:0.3f}".format(
ms(y_train, (y_train_pred))))
print("The MSE score on the Test set is:\t{:0.3f}".format(
ms(y_test, (y_test_pred))))
def ARIMA_(split_len):
data = pd.read_csv("~/Downloads/sunspot.csv")
time = data["time"]
data = data.set_index("time")
train_data = data.iloc[:split_len, :]
test_data = data.iloc[split_len:, :]
model_arima = ARIMA(data, order=(9, 0, 0))
model1 = model_arima.fit(disp=0)
residuals1 = pd.DataFrame(model1.resid)
residuals1.plot()
residuals1.plot(kind='kde')
#plt.show()
#print(residuals1.describe())
#print(model1.summary())
y_train_pred = model1.predict(start=0, end=split_len - 1)
y_test_pred = model1.predict(start=split_len, end=288)
y_actual = np.array(data["value"])
error_train = y_actual[:split_len] - np.array(y_train_pred)
error_test = y_actual[split_len:] - y_test_pred
# one step forecastvalue"
print("MSE Train :", ms(train_data, y_train_pred))
print("MSE Test :", ms(test_data, y_test_pred))
#print(len(error_train),len(y_train_pred))
return error_train, error_test, data
def ARIMA_(split_len):
data = pd.read_csv("~/Downloads/lynx.csv")
time = data["time"]
data = data.set_index("time")
train_data = data.iloc[:split_len, :]
test_data = data.iloc[split_len:, :]
model_arima = ARIMA(data, order=(12, 0, 0))
model1 = model_arima.fit(disp=0)
residuals1 = pd.DataFrame(model1.resid)
residuals1.plot()
residuals1.plot(kind='kde')
plt.show()
print(residuals1.describe())
print(model1.summary())
# one step forecast
y_test_pred = model1.forecast()
y_train_pred = model1.predict(start=0, end=split_len - 1)
y_test_pred = model1.predict(start=split_len, end=114)
print("MSE Train :", ms(train_data, y_train_pred))
def rms(o, p):
return sqrt(ms(o,p))
learning_rate=0.05,
max_depth=3,
n_estimators=2500,
reg_alpha=0.4640,
reg_lambda=0.8571,
random_state=7)
model.fit(X_train, y_train)
print(model)
score_train = model.score(X_train, y_train)
score_test = model.score(X_test, y_test)
output = model.predict(X_test)
score_r2_pred = r2_score(y_test, output)
final_df = pd.DataFrame()
mse = ms(y_test, output)
rmse = sqrt(mse)
Obsv_tbl += [['XGB Regressor', score_train, score_test, score_r2_pred, rmse]]
#ADABOOST REGRESSION MODEL
from sklearn.ensemble import AdaBoostRegressor
parameters = {}
model = GridSearchCV(AdaBoostRegressor(base_estimator=None,
learning_rate=0.05,
loss='linear',
n_estimators=500,
random_state=None),
parameters,
cv=5)