Bike-sharing forecasting: a forecasting 30 minutes ahead without using 'progressive_val_score'

[TxusLopez]

Hi,

I'm using the progressive_val_score function to evaluate the predictions X minutes ahead in my dataset. This function manages this issue by using the delay and moment parameters. I'm not sure about the results I am obtaining. My features and labels are not shifted, so I expected the function to do that, by managing which features at time t should be trained with the corresponding label, etc.

I would like to see how I can do that manually or traditionally without using the function. For example in the Bike-sharing forecasting toy example.

Thanks

[MaxHalford]

Hey! Sorry but what's the question?

I'm asking this in a tongue in cheek manner because this is a complex topic. Do you want to see in what order the data is processed by the model? Under the hood, evaluate.progressive_val_score uses the stream.simulate_qa function. This is the function that manages the data so as to process in the correct order.

[TxusLopez]

Thanks Max, I will check the stream.simulate_qa and I will try to simulate its operation for my own purposes. Thanks a lot.

[MaxHalford]

Feel free to shout out if you need a hand :)