tf.data.experimental.make_csv_dataset: ValueError: `label_name` provided must be one of the columns

Question

I am trying to construct a dataset to use in Keras for the Titanic example on Kaggle. Here is what I've done so far:

train_data = pd.read_csv("/kaggle/input/titanic/train.csv")

all_columns = ['PassengerId', 'Survived', 'Pclass', 'Name', 'Sex', 'Age', 'SibSp', 'Parch', 'Ticket', 'Fare', 'Cabin', 'Embarked'] # all the columns names present in the csv

feature_columns = ['Pclass', 'Sex', 'Age', 'SibSp', 'Parch', 'Ticket', 'Fare', 'Cabin', 'Embarked'] # columns that I want to use as features for the training part


train_data = tf.data.experimental.make_csv_dataset(
    "/kaggle/input/titanic/train.csv",
    batch_size=12,
    column_names=all_columns,
    select_columns=feature_columns,
    label_name='Survived', # name of the 'label' column
    na_value="?",
    num_epochs=1,
    ignore_errors=False)

But when compiling, I get this error:

495   if label_name is not None and label_name not in column_names:
496     raise ValueError("`label_name` provided must be one of the columns.")
497 
498   def filename_to_dataset(filename):

ValueError: label_name provided must be one of the columns.

But, as you can see label_name value is 'Survived' and it's present in all_columns (there also to column_names)

Any idea ?

Best

Aymeric

Answer 1

label_name must be included in the select_columns

Try:

train_data = tf.data.experimental.make_csv_dataset(
    "/kaggle/input/titanic/train.csv",
    batch_size=12,
    column_names=all_columns,
    select_columns=feature_columns + ['Survived'],
    label_name='Survived', # name of the 'label' column
    na_value="?",
    num_epochs=1,
    ignore_errors=False)