I have daily time-series data, which tells me the rain fall & foot fall at a certain shop on that day. Now, I want to predict the foot fall at time $t$, given the previous $2$ observations.

As I'm dealing with time-series data, I thought I could use a RNN, feeding in the previous $2$ observations.

Now, I want it to learn the dependency between rainfall & footfall (i.e, if it's raining, there will be less footfall), and I want it to be able to look at previous rainfall values in order to gauge the current rainfall.

Let's just consider one observation for the time being.

Let $r_t$ be the rain value at time step $t$ and $y_t$ be the footfall at time $t$, $y_t$ is what I want to predict.

I thought I could construct an input like:

$$ [[r_{t-2}, y_{t-2}],\\ [r_{t-1}, y_{t-1}]] $$

in order to predict $y_t$. But, given I'm at timestep $t$ and I know the rainfall $r_t$, it seems like the RNN has no way of accessing this information? If I know it is raining at timestep $t$, then how do I feed the model this?

I have had a look at parallel series but these aren't really what I'm looking for, as I'm using the previous $y_t$ as a feature here essentially.

Is there a way of structuring this to give me what I want?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Browse other questions tagged or ask your own question.