Lead time calculation in inference buffers #70
Description
Hello, I am try to work out what the correct way is to calculate lead time in output rollouts. As far as I can tell, there are two parameters, which I interpreted as follows:
dt: number of dataset timesteps ahead the model predicts; set to 1 if the model predicts consecutive samples from the datasetdhours: number of hours between timesteps in a dataset; set to 6 for fine-tuning/released FCN3 model weights
If my interpretation above is correct, I would expect the rollout_dt argument to the inference buffers to be dt * dhours. As it is, it looks like it’s only dt, leading to incorrect coordinates in the output. Does this need a fix? Happy to make a PR to update inferencer.py.
Maybe @azrael417