# K-Fold CV
kf = KFold(n_splits=2, random_state=None, shuffle=False)
# Load data and label
X, y = load_data()
acc = []
for i in range(10):
acc_local = [] # To store accuracy of k times from k-Fold
for train_index, test_index in kf.split(X):
# Load train and test data and label
train, train_label, test, test_label = X[train_index], y[
train_index], X[test_index], y[test_index]
# Param: X,Y,epoch=100,lr=0.0005
net.fit(train, train_label, epoch=1, lr=0.0005)
acc_fold = 0 # Counter for correct prediction
for j in range(len(test)):
to_feed = torch.tensor(test[j],
dtype=torch.float64,
requires_grad=True).cuda()
pred = net(to_feed.float())[0]
acc_fold += int((pred[0] < pred[1]) == int(
test_label[j])) # Accumulate # of correct prediction
# Log accuracy of each fold
acc_local.append(acc_fold / len(test))
acc.append(np.average(acc_local)) # Accuracy of 1 epoch
for i in dpp.outlier_dict.keys():
dpp.DataStates[i] = Data(dpp.outlier_treatment(dpp.DataStates['Standardized_data'].df,forecastMeasure,outlier_type=i),forecastMeasure,m_count,m_perc,i_count,i_perc)
dpp.imputation(dpp.DataStates[i].df,forecastMeasure,outlier_type=i)
l = set(dpp.DataStates.keys()) - set(['Standardized_data','Normal_Distribution','boxplot'])
d = {'Dataset':[],'Predictions':[]}
for i in range(n_preds):
d['t+'+str(i+1)+' RMSE'] = list()
for i in l:
rnn = RNN(dpp.DataStates[i].df.copy(),forecastMeasure,n_lag,n_preds,nvars)
x, y, scaler = rnn.prepare_data(dpp.DataStates[i].df.copy())
x_train,x_test = rnn.train_test_split(x,test_size=0.3)
y_train,y_test = rnn.train_test_split(y,test_size=0.3)
model = rnn.fit([10,10],x_train, y_train)
forecasts = rnn.forecast(model, x_test)
forecasts = rnn.inverse_transform(forecasts, scaler)
actual = rnn.inverse_transform(y_test.values, scaler)
rmse = rnn.evaluate_forecasts(y_test.values, forecasts)
d['Dataset'].append(i)
for i in range(n_preds):
d['t+'+str(i+1)+' RMSE'].append(rmse[i])
d['Predictions'].append(forecasts)
fdf = pd.DataFrame(d)
n = fdf[fdf['t+1 RMSE']==min(fdf['t+1 RMSE'])]['Dataset'].item()
rnn = RNN(dpp.DataStates[n].df.copy(),forecastMeasure,n_lag,n_preds,nvars)
x, y, scaler = rnn.prepare_data(dpp.DataStates[n].df.copy())
x_train,x_test = rnn.train_test_split(x,test_size=0.7)
y_train,y_test = rnn.train_test_split(y,test_size=0.7)
