Time2Energy/main.py at main · pnv2003/Time2Energy · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
import argparse
import numpy as np
import pandas as pd
from sklearn.preprocessing import MinMaxScaler
import torch

from data import download_data, prepare_data, prepare_loader
import dl.model as dl_model
import dl.train as dl_train
from eda import create_time_features
from eval import plot_actual_forecast, plot_future_forecast
import ml.model as ml_model
import stats.model as stat_model
import stats.transform as T

# hyperparameters
xgb_n_estimators = 1000
xgb_max_depth = 6
xgb_learning_rate = 0.3
xgb_objective = 'reg:squarederror'
xgb_early_stopping_rounds = 50

nn_batch_size = 32
nn_num_epochs = 20
nn_early_stop_patience = 5
nn_early_stop_delta = 0.00001
nn_learning_rate = 0.001
nn_window = 30

device = 'cuda' if torch.cuda.is_available() else 'cpu'

class TimeSeriesScaler:

    def __init__(self):
        self.ts = None

    def fit_transform(self, X):
        self.ts = X
        return T.difference(T.log_transform(X))

    def inverse_transform(self, X):
        return T.inverse_log_transform(T.inverse_difference(X, self.ts))

def main():

    parser = argparse.ArgumentParser(description='Model training and testing')
    subparsers = parser.add_subparsers(dest='command', required=True)

    train_parser = subparsers.add_parser('train', help='Train a model')
    train_parser.add_argument('--model', type=str, required=True, choices=['arima', 'sarima', 'xgb', 'nn'], help='Model to train')

    test_parser = subparsers.add_parser('test', help='Test a model')
    test_parser.add_argument('--model', type=str, required=True, choices=['arima', 'sarima', 'xgb', 'nn'], help='Model to test')

    forecast_parser = subparsers.add_parser('forecast', help='Forecast using a model')
    forecast_parser.add_argument('--model', type=str, required=True, choices=['arima', 'sarima', 'xgb', 'nn'], help='Model to forecast')
    forecast_parser.add_argument('--steps', type=int, default=24, help='Number of steps to forecast')

    args = parser.parse_args()

    if args.command == 'train':
        download_data()

        if args.model == 'arima':
            scaler = TimeSeriesScaler()
            train, test = prepare_data(scaler=scaler, method='none')
            order, seasonal_order = stat_model.arima_params(train)
            model = stat_model.arima_fit(train, order, seasonal_order)
            stat_model.arima_save(model, 'models/arima.pkl')

        elif args.model == 'sarima':
            scaler = TimeSeriesScaler()
            train, test = prepare_data(scaler=scaler, method='none')
            order, seasonal_order = stat_model.arima_params(train)
            model = stat_model.sarima_fit(train, order, seasonal_order)
            stat_model.arima_save(model, 'models/sarima.pkl')

        elif args.model == 'xgb':
            X_train, y_train, X_test, y_test = prepare_data(method='feature')
            params = {
                'n_estimators': xgb_n_estimators,
                'max_depth': xgb_max_depth,
                'learning_rate': xgb_learning_rate,
                'objective': xgb_objective,
                'early_stopping_rounds': xgb_early_stopping_rounds
            }
            model = ml_model.xgb_fit(X_train, y_train, X_test, y_test, params)
            ml_model.xgb_save(model, 'models/xgb.pkl')

        elif args.model == 'nn':
            scaler = MinMaxScaler()
            X_train, y_train, X_test, y_test = prepare_data(scaler=scaler, method='shift', window=nn_window)
            loaders = prepare_loader(X_train, y_train, X_test, y_test, batch_size=32, device=device)

            model = dl_model.CNN_LSTM().to(device)
            criterion = torch.nn.MSELoss()
            optimizer = torch.optim.Adam(model.parameters(), lr=nn_learning_rate)
            model = dl_train.train_model(
                loaders,
                model,
                criterion,
                optimizer,
                device,
                num_epochs=nn_num_epochs,
                early_stop_patience=nn_early_stop_patience,
                early_stop_delta=nn_early_stop_delta
            )
            dl_train.save_model(model, 'models/nn.pth')

    elif args.command == 'test':

        if args.model == 'arima':
            scaler = TimeSeriesScaler()
            train, test = prepare_data(scaler=scaler, method='none')
            model = stat_model.arima_load('models/arima.pkl')
            forecast = stat_model.arima_forecast(model, len(test))
            forecast = scaler.inverse_transform(forecast)
            plot_actual_forecast(test, forecast)

        elif args.model == 'sarima':
            scaler = TimeSeriesScaler()
            train, test = prepare_data(scaler=scaler, method='none')
            model = stat_model.arima_load('models/sarima.pkl')
            forecast = stat_model.arima_forecast(model, len(test))
            forecast = scaler.inverse_transform(forecast)
            plot_actual_forecast(test, forecast)

        elif args.model == 'xgb':
            X_train, y_train, X_test, y_test = prepare_data(method='feature')
            model = ml_model.xgb_load('models/xgb.pkl')
            forecast = ml_model.xgb_predict(model, X_test)
            plot_actual_forecast(y_test, forecast)

        elif args.model == 'nn':
            scaler = MinMaxScaler()
            X_train, y_train, X_test, y_test = prepare_data(scaler=scaler, method='shift', window=30)
            loaders = prepare_loader(X_train, y_train, X_test, y_test, batch_size=32, device=device)

            model = dl_model.CNN_LSTM().to(device)
            model = dl_train.load_model(model, 'models/nn.pth')
            model.eval()
            forecast = model(torch.tensor(X_test, dtype=torch.float32).to(device)).cpu().detach().numpy().flatten()
            forecast = scaler.inverse_transform(forecast.reshape(-1, 1)).flatten()
            plot_actual_forecast(y_test, forecast)

    elif args.command == 'forecast':

        if args.model == 'arima':
            scaler = TimeSeriesScaler()
            train, test = prepare_data(scaler=scaler, method='none')
            model = stat_model.arima_load('models/arima.pkl')
            forecast = stat_model.arima_forecast(model, args.steps)
            forecast = scaler.inverse_transform(forecast)
            plot_future_forecast(train, forecast)

        elif args.model == 'sarima':
            scaler = TimeSeriesScaler()
            train, test = prepare_data(scaler=scaler, method='none')
            model = stat_model.arima_load('models/sarima.pkl')
            forecast = stat_model.arima_forecast(model, args.steps)
            forecast = scaler.inverse_transform(forecast)
            plot_future_forecast(train, forecast)

        elif args.model == 'xgb':
            X_train, y_train, X_test, y_test = prepare_data(method='feature')
            model = ml_model.xgb_load('models/xgb.pkl')

            new_index = pd.date_range(X_test.index[-1], periods=args.steps+1, freq='H')[1:]
            X_future = create_time_features(pd.DataFrame(index=new_index), split=False, label=False)
            y_future = ml_model.xgb_predict(model, X_future)
            y_test_sample = y_test.iloc[-args.steps:]
            plot_future_forecast(y_test_sample, y_future)

        elif args.model == 'nn':
            scaler = MinMaxScaler()
            X_train, y_train, X_test, y_test = prepare_data(scaler=scaler, method='shift', window=30)
            loaders = prepare_loader(X_train, y_train, X_test, y_test, batch_size=32, device=device)

            model = dl_model.CNN_LSTM().to(device)
            model = dl_train.load_model(model, 'models/nn.pth')

            X_future = X_test[-nn_window:].values
            X_future_tensor = torch.tensor(X_future, dtype=torch.float32).to(device)
            y_future = []

            model.eval()
            with torch.no_grad():
                for i in range(args.steps):
                    y_pred = model(X_future_tensor.reshape(1, -1))
                    y_future.append(y_pred.item())
                    X_future = np.roll(X_future, -1)
                    X_future[-1] = y_pred.item()
                    X_future_tensor = torch.tensor(X_future, dtype=torch.float32).to(device)

            y_future = scaler.inverse_transform(np.array(y_future).reshape(-1, 1)).ravel()
            y_test_sample = y_test.iloc[-args.steps:]
            y_test_sample = scaler.inverse_transform(y_test_sample.values.reshape(-1, 1)).ravel()
            plot_future_forecast(y_test_sample, y_future)

if __name__ == '__main__':
    main()