I am new to machine learning and using tf/keras to build a simple model to train on Adult database which I downloaded from UCI.

The data has 14 features. I am using keras FeatureSpace API to contruct a DNN model. However, whenever I use more than 3 features, I get following error. It doesn’t matter what 3 features I use.

2024-07-17 22:24:38.439243: W tensorflow/core/framework/op_kernel.cc:1817] OP_REQUIRES failed at cast_op.cc:122 : UNIMPLEMENTED: Cast int32 to string is not supported

….

Cast int32 to string is not supported
[[{{node functional_1/Cast_2}}]] [Op:__inference_one_step_on_iterator_82943]

I initially thought that the data might be incorrect and might have int32 in string column, but I have seen all the unique values for all the columns and that doesn’t seem to be the case.

I also tried different combinations of features and as long as these are three features, it works fine but once I add the fourth feature it gives me that error.

Here is my code to create the model and training it:

def construct_feature_space():
  feature_space = tf.keras.utils.FeatureSpace(
    features = {
      'age': tf.keras.utils.FeatureSpace.float_discretized(10),
      'workclass': 'string_categorical',
      'education_num': tf.keras.utils.FeatureSpace.float_discretized(10),
      'race': 'string_categorical',
      #'native_country': 'string_categorical',
      #'occupation': 'string_categorical'
    }
  )
  return feature_space

def create_fs_model(training_dataset):
  feature_space = construct_feature_space()
  feature_space.adapt(training_dataset)

  inputs = feature_space.get_inputs()
  encoded_features = feature_space.get_encoded_features()

  h1 = tf.keras.layers.Dense(128, activation="relu")(encoded_features)
  outputs = tf.keras.layers.Dense(1, activation='sigmoid', name="high_income")(h1)
  
  model = tf.keras.Model(inputs, outputs)
  model.compile(optimizer='adam',loss='mse',metrics=['accuracy'])
  print(model.summary())
  tf.keras.utils.plot_model(model, 'dnn_model.png', show_shapes=False, rankdir='LR')
  return model

model = create_fs_model(train_raw_dataset)
model.compile(optimizer='adam',loss='binary_crossentropy',metrics=['accuracy'])

history = model.fit(train_dataset,
          validation_data=eval_dataset,
          epochs=10,
          steps_per_epoch=2)