So far for TensorBoard Projector, for input to the sausage makers (UMAP, t-SNE, PCA) we've only been using the raw pixels. But along with those can be added "metadata" for filtering. We really want to get that data off the blockchain and into Projector
The very first bit could be "is it for sale?" so a person could filter by that binary value. And then all the attributes as defined in the NFTs themselves.
(Note: what would be really cool is if we were to simply reduce the transparency of those that are not for sale. This would still provide the hyperdimensional neighborhood context but let shoppers focus on what is available. But such is not in the code of Projector. Maybe we could set the transparency before sending the images in. Not sure if Projector would honor transparency but I bet it would. Even if Projector cannot do this, us doing this would be a great demo. For a productified implementation we would either need to be constantly rebuilding the tranparents, or there would need to be our own code wrapping Projector to do this live.)
Another filtering by attribute would be: isTokenInWallet. This would be another use case for increasing transparency – of all tokens not in the wallet with id = foo.
So far for TensorBoard Projector, for input to the sausage makers (UMAP, t-SNE, PCA) we've only been using the raw pixels. But along with those can be added "metadata" for filtering. We really want to get that data off the blockchain and into Projector
The very first bit could be "is it for sale?" so a person could filter by that binary value. And then all the attributes as defined in the NFTs themselves.
(Note: what would be really cool is if we were to simply reduce the transparency of those that are not for sale. This would still provide the hyperdimensional neighborhood context but let shoppers focus on what is available. But such is not in the code of Projector. Maybe we could set the transparency before sending the images in. Not sure if Projector would honor transparency but I bet it would. Even if Projector cannot do this, us doing this would be a great demo. For a productified implementation we would either need to be constantly rebuilding the tranparents, or there would need to be our own code wrapping Projector to do this live.)