More outputs then labels

Research & Models
#1

Hey I want to create a model what outputs four 3-dimensional Vectors [batch, time, feature]. One of these output is just for visualization of some model-intern behavior and should not be accounted in metrics or losses. Resulting of this, i have more modelloutputs than ground-truths. This leads to errors. ( ValueError: The two structures don’t have the same sequence length. Input structure has length 3, while shallow structure has length 4.) How can i solve this?

#2

Hi @Samuel_K ,

Here are two possible ways to solve the error caused by having more model outputs than ground-truths:

  1. Define a custom loss function that takes into account only the first three outputs for calculating metrics and losses. This way, the fourth output, which is meant only for visualization, will be ignored.
  2. Alternatively, you could use a masking layer to mask out the fourth output. This would ensure that the number of model outputs and ground-truths match, avoiding the error.

I hope this helps you!

Thanks