I want to visualize a time series of a single high dimensional tensor (like an embedding vector) in the tensorboard projector. So I want to visualize all previous versions of this tensor in one visualization, like rows in an embedding matrix, and maybe associate some metadata with each version (e.g. time step and loss at this time step). Is there a possibility to do this? The only way I could imagine is to store all versions of the tensor and then visualize them like a normal embedding matrix. But how to do so?
Thanks for any help!
For choosing different checkpoints for visualization in the projector, I found this: How can I select which checkpoint to view in TensorBoard's embeddings tab? But that's not what I looking for.
