I have many products in my warehouses which can be "demanded" any day by my different clients. I want to forecast how many of each item will be demanded for the whole next year. Naturally, as I'm having a large amount of different time series (each one for a different product, with potentially some shared patterns and correlations between them), I want to train some kind of RNN (could be a simple LSTM or even something more complex like Amazons DeepAR or Googles TemporalFusionTransformer). Also, as the data is unevenly spaced (there can be days or even months without any demand), I'm planning to provide the model with a timestamp delta (dts) as an additional input feature.

The problem is, how can I use such trained network to predict the total demand for one year ahead? As the network trained with unevenly spaced data, I cant determine the time window for each new prediction, let alone a full year. How can I tweak the network to achieve this? Or maybe I am tackling the problem from a wrong direction? (I did try a couple of different solutions like Prophet who have been fruitless)

Thanks in advance.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.