Issue with training clasification Neural Network

Question

I have converted a regression machine learning problem to a clasification one via the pd.cut() function (basically splits the regression range of the labels into bins). When trying to train the model I get various issues such as

 ValueError                                Traceback (most recent call last)
<ipython-input-93-bea8c9e6a3db> in <module>()
     18 
     19 # 3. Fit the model
---> 20 model_10.fit(X_train, y_train, epochs=200)
     21 # , validation_data=(X_test, y_test)

1 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
    100       dtype = dtypes.as_dtype(dtype).as_datatype_enum
    101   ctx.ensure_initialized()
--> 102   return ops.EagerTensor(value, ctx.device_name, dtype)
    103 
    104 

ValueError: Failed to convert a NumPy array to a Tensor (Unsupported object type pandas._libs.interval.Interval).

I tried converting the data to tensor format or numpy arrays, but a Graph error appeared:

In this image you can see how the labels of the model look

# Set seed
tf.random.set_seed(18)

# 1. Create model
model_10= tf.keras.Sequential([
                                      
                                      tf.keras.layers.Dense(50, activation='relu'),
                                      tf.keras.layers.Dense(25, activation='relu'), 
                                      tf.keras.layers.Dense(25, activation='relu'),
                                      tf.keras.layers.Dense(15, activation=tf.keras.activations.softmax)

])

# 2. Compile model
model_10.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(),
                 optimizer=tf.keras.optimizers.SGD(lr=0.01),
                 metrics=['accuracy'])

# 3. Fit the model
model_10.fit(X_train, y_train, epochs=200)

above is the model I tried to use, (the data was split with sklearn)

I also tried to simply round the regression values to the nearest int (5 classes in total). Nonetheless this implementation also didn't prove to work.