My input features are 1d arrays of shape (1000,)
I can tokenize the arrays using tf.extract_patches
block_size = self.token_shape[0]
stride = self.token_shape[0]
tokens = extract_patches(images = reshaped, sizes=[1, 1, block_size, 1],
strides=[1, 1, stride, 1], rates=[1, 1, 1, 1], padding='VALID')
However, I'm having trouble with using the tokens as input to the SWIN transformer.