Reshaping spatiotemporal dimension of mmap numpy so that it can be fed to ConvLSTM2D I have this tensor of multidimensional array:
