Autokeras StructuredDataClassifier InvalidArgumentError: Graph execution error

Question

I'm using Auokeras to choose a model for my dataset. I've tried using StructuredDataRegressor and everything went ok. But now I'm trying to use StructuredDataClassifier and can't get it running:

df =

   edad target  F   M
13  62  0.75    1.0 0.0
39  56  0.75    1.0 0.0
30  49  0.75    1.0 0.0
45  73  1.00    1.0 0.0
17  67  0.75    0.0 1.0
48  57  1.00    1.0 0.0
26  71  1.00    0.0 1.0
....

I'm doing:

train_set, test_set = train_test_split(df, test_size=0.2, random_state=42)
test_set
x_train = train_set
y_train = x_train.pop("target")
x_test = test_set
y_test = x_test.pop("target")
reg = ak.StructuredDataClassifier(max_trials=3, overwrite=True)
reg.fit(x_train, y_train, epochs=3)

But only with the Classifier I get:

InvalidArgumentError: Graph execution error:

indices[0] = -1 is not in [0, 4)
     [[{{node embedding_lookup}}]]
     [[IteratorGetNext]] [Op:__inference_adapt_step_30457]

Tried changing my target variable to Integer. I've tried with other datasets (titanic) and everything is ok, but I do not see the difference with mine. I've seen this error in the web but i do not understand why it doesn't show up when I use the Resgressor.

Thank's in advance.