I wonder how I can output only the next time step forecast and record it and use the actual value at that time step to forecast the value at the next time step? All the recorded forecasts can then tell me how accurate my model is.
To do naive forecast
, you can use a loop construct in python and call model.fit
for custom batches. Please keep 2 things in mind when it comes to model training:
- If you call
model.fit
, say, 2 times,then, it’s the same as training the model using 2 batches of data. - A batch of xs and ys is sufficient to invoke
model.fit
. It’s ok to come up with this batch on the fly.