What is the relationship between the arg "periods" in make_future_dataframe() and "n_forecasts" in model

[HamletMonkey]

I created a model m train on daily data from Jan-Jun20.

m = NeuralProphet(
    growth = "off",
    yearly_seasonality=False,
    weekly_seasonality=False,
    daily_seasonality=False,
    n_lags=2,
    learning_rate=0.003,
)

I am trying to use m.make_future_dataframe() to forecast 3 months ahead (Jul-Sep20). However, passing in different value for the periods argument does not change the number of future forecast - I'm only getting the date 2020-07-01 regardless. Am I missing out something here :( ?

Adding in n_forecasts=30 to the model gives me the 3-months future dataframe (Jul-Sep20), but am still struggling to wrap my head around the Prediction Collection concept of yhat(s) here, and are they even related?

[poo5297]

Did you find out an explaination to this?

[SeanKski]

I'm also struggling with this...

[skant626]

I have the same question...

[mmangione]

You guys also need to specify how many historic predictions you are going to use.

My code for this looks like:
future = m.make_future_dataframe(observed_df, periods=n_days_forecasts, n_historic_predictions=len(observed_df))

FYI: n_historic_predictions >= n_lags.

[sdegeye]

I created a model m train on daily data from Jan-Jun20.

m = NeuralProphet(
    growth = "off",
    yearly_seasonality=False,
    weekly_seasonality=False,
    daily_seasonality=False,
    n_lags=2,
    learning_rate=0.003,
)

I am trying to use m.make_future_dataframe() to forecast 3 months ahead (Jul-Sep20). However, passing in different value for the periods argument does not change the number of future forecast - I'm only getting the date 2020-07-01 regardless. Am I missing out something here :( ?

Adding in n_forecasts=30 to the model gives me the 3-months future dataframe (Jul-Sep20), but am still struggling to wrap my head around the Prediction Collection concept of yhat(s) here, and are they even related?

[sdegeye]

Totally agree with you, the documentation give the focus on the possible plots but the way to manage a long term prediction is not clear.
The process is very slow. I expected that even if we specify a n_forecasts argument of for instance 10 and we need to forecast a complete dataframe of 1000 records in the future that the fonction manage itself the repetition n times of the n_forecast to complete the entire dataframe predictions but it is not the case or maybe I'm wrong? If we have to manage ourself a loop to achieve that, that's not really super.
What's your opinion about that?