Autoregression n_lags causes null predictions on basic dataset

[suntochukwu]

Adding autoregression via n_lags causes results to be null?? The data set is just gold stock market prices- ds/datetime and closing price/y. Here is the code for the model:

m2 = NeuralProphet(drop_missing= True , growth= 'discontinuous', n_lags=50)
m2.set_plotting_backend("plotly-static")
metrics = m2.fit(basicdf, freq= 'H')
#future = m2.make_future_dataframe(basicdf, periods=50, n_historic_predictions=True)
forecast = m2.predict(future)

Without n_lags the models works and predicts well. When i add autoregression i get an error:

the prediction works but all the yhat1 values are NaN. Any ideas: the model should be smart enough to know the first few values are lagged, also why would it not predict anything when the n_lag is small??? Remove the n_lags and this works fine. I feel like this is a very basic feature to be bugging.

Dataset:
ds step0 trend0 season_weekly0 season_daily0 ar0 step1 trend1 season_weekly1 season_daily1 ... step98 trend98 season_weekly98 season_daily98 ar98 step99 trend99 season_weekly99 season_daily99 ar99
0 2023-06-08 20:00:00 NaN NaN NaN NaN NaN NaN NaN NaN NaN ... NaN NaN NaN NaN NaN NaN NaN NaN NaN NaN
1 2023-06-08 21:00:00 NaN NaN NaN NaN NaN NaN NaN NaN