Problem with sample_weight on tf.keras: “sample_weight cannot be broadcast”

Question

I am working on a Recurrent Neural Network and with three layers and a few units per cell, say 100. I have trained with success a network where I only care about the last ou