Time series forecasting with different interval (input and prediction)

Hi everyone,

i am new to fast.ai and this is my first post .so forgive if it seems stupid question .

i have data with hourly interval and i want to forecast for every 30 mins going forward , how should i proceed with this . i do not want to change my input sequence from hourly to half hourly by averaging .
looking for all your guidance .thanks .

In supervised learning you have to have examples (best loads of them) of the inputs and the expected outputs of the model you want to create. The model learns by seeing the input and the expected correct output.

So - if you want to predict 30min intervals you need to have timeseries with 30min intervals. You could then downsample them to 1h intervals and do training with those 2 timeseries. But there is no model that can sensibly “invent” higher resolution timeseries without first being shown both.

As you write you have 1h timeseries, what you are trying to do will be very difficult if not impossible.

thanks Marc for suggestion, i have used facebook’s prophet and changed frequency parameter for 30 mins ,without changing any input sequence .it outputs 30 mins interval predicted.

and also i upsample data from 1 hr to 30 mins and input it into RNN -LSTM .

okay, thanks for pointing out fb prophet, I had never heard of that. Seems very interesting for regular time series data (but has nothing to do with deep learning, they do curve fitting, which is why the interval of the time series doesn’t matter that much in that case).