CODEWITHSUNDEEP
pythontensorflowdeep-learningvalueerror
Florian
Florian

Reputation: 89

Tensorflow: ValueError: Expected non-integer, got <dtype: 'int32'>

I'm just starting with Tensorflow and when I call m.fit(input_fn=lambda: self.input_fn(train_data), steps=train_steps), then I receive the following error.

File "/Library/Python/2.7/site-packages/tensorflow/contrib/layers/python/layers/feature_column_ops.py", line 161, in _input_from_feature_columns
    transformed_tensor = transformer.transform(column)
File "/Library/Python/2.7/site-packages/tensorflow/contrib/layers/python/layers/feature_column_ops.py", line 882, in transform
    feature_column.insert_transformed_feature(self._columns_to_tensors)
File "/Library/Python/2.7/site-packages/tensorflow/contrib/layers/python/layers/feature_column.py", line 991, in insert_transformed_feature
    self.sparse_id_column.insert_transformed_feature(columns_to_tensors)
File "/Library/Python/2.7/site-packages/tensorflow/contrib/layers/python/layers/feature_column.py", line 572, in insert_transformed_feature
    name="lookup")
File "/Library/Python/2.7/site-packages/tensorflow/contrib/lookup/lookup_ops.py", line 1018, in index_table_from_tensor
    "integer" if dtype.is_integer else "non-integer", keys.dtype))
ValueError: Expected non-integer, got <dtype: 'int32'>.

In the feature columns that I pass to fit(), there are only int32and int64, but that should not be the problem, should it?

Upvotes: 0

Views: 2361

Answers (1)

Yury Kozyrev
Yury Kozyrev

Reputation: 68

I think it could happen that you use categorical features with tf.SparseTensor but your feature columns contain int32.

In this case just convert your integer columns into string, for example like this:

# using Pandas
for f in categorical_features:
    df_train[f] = df_train[f].astype(str)   
    df_test[f] = df_test[f].astype(str)

Upvotes: 2

Related Questions