business-science / modeltime.gluonts

GluonTS Deep Learning with Modeltime
https://business-science.github.io/modeltime.gluonts/
Other
39 stars 9 forks source link

Error when tuning N-BEATS: lookback_length and scale #19

Open vidarsumo opened 3 years ago

vidarsumo commented 3 years ago

I'm tuning N-BEATS and I get an error when trying to tune the lookback_length. i Creating pre-processing data to finalize unknown parameter: lookback_length Error: The workflow has arguments whose ranges are not finalized: 'lookback_length'

I can solve this by using update:

tune_nbeats_results <- tune_grid(
    object     = wflw_nbeats,
    resamples  = resamples_tscv,
    param_info = parameters(wflw_nbeats) %>% 
        update(lookback_length = lookback_length(range = c(1, 7) * horizon)),
    grid       = 5,
    control    = control_grid(verbose = TRUE)
)

Is there any way for this to be automatic so we don't have to use the update function?

I also get an error when trying to tune the scale hyperparameter: Error: Problem with mutate() input object. x Error when calling scale(): Error : 'scale' is not an exported object from 'namespace:modeltime.gluonts'

I saw that you renamed the function, scale_values() but I also get an error if change scale to scale_values inside nbeats(...): unused argument (scale_values = tune())