Dimensions must be Equal, but are 3 and 5 for...with input shapes: [32,3], [32,5]

Question

I’m building a neural network, and I find the following mistake:

ValueError: Dimensions must be equal, but are 3 and 5 for '{{node mean_absolute_error/sub}} = Sub[T=DT_FLOAT](sequential/dense_1/Sigmoid, IteratorGetNext:1)' with input shapes: [32,3], [32,5].

Since I have used other bases in the same code and it worked. And the bases are arranged in the same way.

I have tried changing the Oss, changing the optimizer or metric, and nothing. I also can’t find anything in the forums.

import pandas as pd
from sklearn.preprocessing import OneHotEncoder
from sklearn.model_selection import train_test_split
import numpy as np


entradas = pd.read_csv('class_in.csv')
saidas = pd.read_csv('class_out.csv')

X = entradas
X = np.array(X)
Y = saidas
Y = np.array(Y)
onehot_encoder=OneHotEncoder(sparse=False)
reshaped=Y.reshape(len(Y), 1)
y_onehot=onehot_encoder.fit_transform(reshaped)
print (X)
print(y_onehot)

callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=50)

from sklearn.model_selection import StratifiedKFold
K = 5
                         
skf = StratifiedKFold(n_splits = K, shuffle = True) 

performance = [None] * (K)
fold = 0

for train_val, test in skf.split(X,Y):
  model = tf.keras.models.Sequential([
  tf.keras.layers.Dense(300, input_dim=200, activation=tf.nn.sigmoid),
  tf.keras.layers.Dense(3, activation=tf.nn.sigmoid)
  ])

  model.compile(optimizer='sgd',
              loss='mae',
              metrics=['accuracy'])

  x_train, x_val, y_train, y_val = train_test_split(X[train_val], y_onehot[train_val], test_size=0.2, stratify=Y[train_val])


  model.fit(x=x_train, y=y_train, epochs=1000000, callbacks=[callback], validation_data=(x_val,y_val))
  performance[fold] = model.evaluate(X[test], y_onehot[test], verbose=0)
  fold = fold + 1