Error in computing_metrics function (incompatible shapes)

Course Q&A TensorFlow Developer Professional Natural Language Processing in TensorFlow
1

Hello
I am facing this error when running the computing_metrics function:

InvalidArgumentError: Incompatible shapes: [361,32] vs. [361] [Op:SquaredDifference]

The function receives this parameters:

mse, mae = compute_metrics(series_valid, rnn_forecast)

It seems “series_valid” and “rnn_forecas” are not having the same shape, but I don`t really understand why. I have not touched the definition of those functions in the notebook.

Thanks,

Boris Martínez

2

Your model output is incompatible with the validation series. Please ensure that your model produces the correct number of features per timestep at the output layer.

Here are a couple of examples to interpret the output shape:

  1. (10, ): model produces 10 timesteps of output with 1 feature per timestep
  2. (10, 32): model produces 10 timesteps of output with 32 features per timestep.
3

Yes, thank you very much. I solved the issue!