MAPE: nan RMSE: nan about darts

[dennisbader]

Hi @SyedKumailHussainNaqvi, this looks like there are some missing values either in your target or predicted series.

[SyedKumailHussainNaqvi]

Thank for your kind guidance, I fix this but i have another query about TransformerModel.
I have these train, validation and test series after scaling now i just predict the trained Transformer model on test series for forecasting without n – The number of time steps after the end of the training time series, how its possible?

scaler = MinMaxScaler(feature_range=(-1, 1))
scaler_target = Scaler(scaler)
scaler_input = Scaler(scaler)
Train_Target_Active_Power_scaled = scaler_target.fit_transform(Target_train_series)
Train_Input_Features_scaled = scaler_input.fit_transform(Input_train_series)

val_Target_Active_Power_scaled = scaler_target.transform(Target_val_series)
val_Input_Features_scaled = scaler_input.transform(Input_val_series)

Test_Target_Active_Power_scaled = scaler_target.transform(Target_test_series)
Test_Input_Features_scaled = scaler_input.transform(Input_test_series)

my_model = TransformerModel(
    input_chunk_length=32,
    output_chunk_length=288,
    batch_size=128,
    n_epochs=1,
    #model_name="air_transformer",
    nr_epochs_val_period=100,
    d_model=32,
    nhead=16,
    num_encoder_layers=8,
    num_decoder_layers=8,
    dim_feedforward=256,
    dropout=0.1,
    activation="relu",
    #random_state=42,
    #save_checkpoints=True,
    force_reset=True,
    pl_trainer_kwargs={
      "accelerator": "gpu",
       "devices": 1
    },
)
my_model.fit(
    series=Train_Target_Active_Power_scaled,past_covariates=Train_Input_Features_scaled, val_series=val_Target_Active_Power_scaled,val_past_covariates=val_Input_Features_scaled,
    verbose=True,
);
pred_series = my_model.predict(n=len(Test_Target_Active_Power_scaled), series=Test_Target_Active_Power_scaled, past_covariates=Test_Input_Features_scaled)

Actual_Pred= scaler_target.inverse_transform(pred_series)
Actual_val_Target_Active_Power= scaler_target.inverse_transform(Test_Target_Active_Power_scaled)

mape_score = mape(Actual_Pred,Actual_val_Target_Active_Power)
mse_score = mse(Actual_Pred,Actual_val_Target_Active_Power)
mae_score = mae(Actual_Pred, Actual_val_Target_Active_Power)
print(f'MAPE: {mape_score:.4f}')
print(f'RMSE: {mse_score:.4f}')
print(f'MAE: {mae_score:.4f}')

[madtoinou]

Hi @SyedKumailHussainNaqvi,

I have these train, validation and test series after scaling now i just predict the trained Transformer model on test series for forecasting without n – The number of time steps after the end of the training time series, how its possible?

I don't understand what you mean by this, your code snippet is clearly relying on n to indicate the length of the forecasts. This argument is mandatory.

If you want to forecast the series with the same index as your test series in order to compute metrics, you need to use my_model.predict(n=len(test_series), series=validation_series, covariates=...) as the prediction time index will start where the input series finishes.