What’s more, this post can serve as an introduction to building your own modules. If you’re interested in quickly playing around with a Keras convLSTM, check out this nice example. For one, we have to – there is no official PyTorch implementation. Instead, we’ll take our time to build a convolutional LSTM (henceforth: convLSTM) in torch. Today, we won’t jump into real-world applications just yet. We want both: recurrence relations and convolutional filters. Importantly, we need to operate in both space and time. We do but, once we feed spatial data to an RNN, treating different locations as different input features, we lose an essential structural relationship.
For time-series prediction, we have that time-honored set of recurrent architectures (e.g., LSTM, GRU), right? Right. Alternatively, it could be a univariate time series, like a meteorological index.īut wait a second, you may be thinking. The target to be predicted could then span that same (or another) grid. By spatially-determined I mean that whatever the quantities we’re trying to predict – be they univariate or multivariate time series, of spatial dimensionality or not – the input data are given on a spatial grid.įor example, the input could be atmospheric measurements, such as sea surface temperature or pressure, given at some set of latitudes and longitudes.
#Height and width layer list stack series
This post is the first in a loose series exploring forecasting of spatially-determined data over time.
0 kommentar(er)
