How to load a model and continue predicting without re-training?

[manujosephv]

Let's say I have a dataset $X = {x_1, ..., x_t}$ and I trained the model with a context window of $k$ to predict next $h$ timesteps.

Now when I do predict, $x_{t-k}, ..., x_t$ is used to predict $x_{t+1},..., x_{t+h}$.

This I can achieve easily using neuralforecast.

Now, I have moved ahead in time to $t+1$ and have the actual observation. I want to use the same saved model, but use the updated window ($x_{t+1-k}, ..., x_{t+1}$), to predict $x_{t+1}, ..., x_{t+h+1}$.

How can I do this in neuralforecast? I assume the dataset would need to be updated in some way so that the last window is including the new datapoint?

[manujosephv]

I took one look at the predict method and realized how stupid the question is!

if df has the newest data, and future_df is the future data,

nf.predict(df=df, futr_df=future_df)

🤦🏼
Feel free to delete the discussion!