In the Flux docs for
train! it says
In case datapoints
dare of numeric array type, assume no splatting is needed and compute the gradient of
What does “splatting” mean here?
I am guessing that what this means is that, if the data are expressed as matrices (dimensions X datapoints) or maybe the other way around, the loss function can be written to operate on multiple datapoints at once. Is this correct?
If so, then does Flux assume that if the data are in a matrix, that the loss must operate on the whole matrix at once?