ValueError: Invalid dtype: str768 about tensorflow

[nlbrown62]

I had a similar issue when I upgraded to Tensorflow 2.16.1, Python 3.12.2. In my case, I triggered the error "ValueError: Invalid dtype: str39968" when trying to run the 'fit' function for a very similar RNN with string data (dtype <U1249). For a quick workaround, the fit function works when setting the dtype to a generic 'object' (i.e. x_train = x_train.astype(object). In Tensorflow 2.15.0 , Python 3.9.16, I didn't have the same issue.

[sushreebarsa]

@maxlelyonais Please make sure the string data uses the same encoding as the rest of your code?
Could you provide the access to the colab gist you have provided?
Thank you!

[maxlelyonais]

Sorry for the delay, yes here is the colab gist: https://colab.research.google.com/gist/maxlelyonais/28bd9f80a71a55ec9513950c057cf5c6/clasificacion_de_spam.ipynb

[sushreebarsa]

@maxlelyonais Could you please have a look at the workaround provided by @nlbrown62 and let us know?
Thank you!

[maxlelyonais]

It works perfectly, May I know why using 'object' works?

[sushreebarsa]

@maxlelyonais object dtype in TensorFlow acts as a catch-all for various data types, including strings of any length.
Thank you!