transfer_learning_all_m3.py

# load M3 dataset
m3_train, m3_test = load_m3()

# naive last
naive1_smapes_m3, naive1_time_m3 = eval_local_model(
    m3_train, m3_test, NaiveSeasonal, K=1
)

# naive seasonal
naive12_smapes_m3, naive12_time_m3 = eval_local_model(
    m3_train, m3_test, NaiveSeasonal, K=12
)

# Exponential smoothing
ets_smapes_m3, ets_time_m3 = eval_local_model(m3_train, m3_test, ExponentialSmoothing)

# Theta
theta_smapes_m3, theta_time_m3 = eval_local_model(m3_train, m3_test, Theta)

# ARIMA
arima_smapes_m3, arima_time_m3 = eval_local_model(
    m3_train, m3_test, ARIMA, p=12, d=1, q=0
)

# Kalman filter
kf_smapes_m3, kf_time_m3 = eval_local_model(
    m3_train, m3_test, KalmanForecaster, dim_x=12
)

# Linear regression
lr_smapes_m3, lr_time_m3 = eval_global_model(
    m3_train, m3_test, LinearRegressionModel, lags=30, output_chunk_length=1
)

# LGBM
lgbm_smapes_m3, lgbm_time_m3 = eval_global_model(
    m3_train, m3_test, LightGBMModel, lags=35, output_chunk_length=1, objective="mape"
)

# Get forecasts with our pre-trained N-BEATS
start_time = time.time()
preds = nbeats_model_m4.predict(series=m3_train, n=HORIZON)
nbeats_m4_elapsed_time_m3 = time.time() - start_time
nbeats_m4_smapes_m3 = eval_forecasts(preds, m3_test)

# Get forecasts with our pre-trained linear regression model
start_time = time.time()
preds = lr_model_m4.predict(series=m3_train, n=HORIZON)
lr_m4_elapsed_time_m3 = time.time() - start_time
lr_m4_smapes_m3 = eval_forecasts(preds, m3_test)

# Get forecasts with our pre-trained LightGBM model
start_time = time.time()
preds = lgbm_model_m4.predict(series=m3_train, n=HORIZON)
lgbm_m4_elapsed_time_m3 = time.time() - start_time
lgbm_m4_smapes_m3 = eval_forecasts(preds, m3_test)