CODEWITHSUNDEEP
pythontensorflow
Wha
Wha

Reputation: 23

Using Tensorflow embedded columns raises All feature_columns must be _FeatureColumn instances error

I am new to tensorflow and I was trying to follow the official documentation where I came across tf.feature_column.categorical_column_with_vocabulary_list

The code I tested is: 

key='colors', vocabulary_list=('X', 'R', 'G', 'B', 'Y'), default_value=0)
columns = [[tfc.embedding_column(colors, 3)], ...]
features = tf.io.parse_example(..., features=tfc.make_parse_example_spec(columns))
dense_tensor = tfc.input_layer(features, columns)

However , when I run this sample code I get the following error : ValueError: All feature_columns must be _FeatureColumn instances. Given: [EmbeddingColumn(categorical_column=VocabularyListCategoricalColumn(key='colors', vocabulary_list=('X', 'R', 'G', 'B', 'Y'), dtype=tf.string, default_value=0, num_oov_buckets=0), dimension=3, combiner='mean', initializer=, ckpt_to_load_from=None, tensor_name_in_ckpt=None, max_norm=None, trainable=True)]

What I am doing wrong?

Upvotes: 1

Views: 298

Answers (1)

user11530462
user11530462

Reputation:

make_parse_example_spec expects FeatureColumn instances. You can create the FeatureColumn instance using the below method for the category list.

colors = feature_column.categorical_column_with_vocabulary_list(key='colors',vocabulary_lis=('R', 'G', 'B', 'Y'),num_oov_buckets=2)
my_feature_columns = [feature_column.indicator_column(colors)]
feature_column.make_parse_example_spec(my_feature_columns)

Output :

{'colors': VarLenFeature(dtype=tf.string)}

If you want to create a dense embedding tensor on your categorical column, you can follow the below example. 

data = {'colors': ['X', 'R', 'G', 'B', 'Y']}

df = pd.DataFrame(data)

colors = feature_column.categorical_column_with_vocabulary_list('colors', df['colors'].unique())

colors_embedding = feature_column.embedding_column(colors, dimension=4)

dense_tensor = tf.keras.layers.DenseFeatures(colors_embedding)(data)

Result: 

tf.Tensor(
[[ 0.17071894  0.29407692 -0.26661882  0.07768019]
 [ 0.26196313  0.14372464 -0.41102907 -0.7207164 ]
 [-0.7888006  -0.07049363 -0.49007863  0.45744416]
 [ 0.56329435 -0.7051675   0.04742934 -0.69377   ]
 [-0.52031726  0.488502   -0.37031132 -0.44338205]], shape=(5, 4), dtype=float32)

Upvotes: 1

Related Questions