zii
zii

Reputation: 35

Tensorflow Estimator API Input Tensor Names

When using an estimator in tensorflow and passing the inputs using tf.estimator.inputs.numpy_input_function(), what are the names of the tensors being created for the features and labels' inputs.

If I print out the name of all the placeholders in my graph I get the following: name: "enqueue_input/Placeholder" name: "enqueue_input/Placeholder_1" name: "enqueue_input/Placeholder_2"

However, the shape of those tensors were not specified so I cannot tell which is which or why there are 3 of them instead of only one features and one labels tensors.

I realize this question was also asked here: TensorFlow: What are the input nodes for tf.Estimator models

But no one answered.

Upvotes: 2

Views: 963

Answers (1)

MatthewScarpino
MatthewScarpino

Reputation: 5936

numpy_input_fn has two important arguments: x and y. x is a dictionary that matches the names of feature columns to arrays containing feature data. y is an array that contains the labels for the features in x.

For example, the following code associates the feature named x_coord with Label 1 when the value is 0.5 and with Label 2 when the value is 1.2:

train_input = tf.estimator.inputs.numpy_input_fn(
    x={"x_coord": np.array([0.5, 1.2])}, y=np.array([1, 2]))

Upvotes: 1

Related Questions