Modification of code to account for different features

[Vincent9797]

Hi,

I am trying to put a tensorflow Estimator class into production. My features are in the form of this:
{
"input_ids":
tf.constant(
features[0].input_ids, shape=[1, 128],
dtype=tf.int32),
"input_mask":
tf.constant(
features[0].input_mask,
shape=[1, 128],
dtype=tf.int32),
"segment_ids":
tf.constant(
features[0].segment_ids,
shape=[1, 128],
dtype=tf.int32),
"label_ids":
tf.constant(features[0].label_id, shape=[1], dtype=tf.int32),
}

When I tried to edit the code for queued_predict_input_fn for the threaded classifier, I am unsure of what to assign output_types to.

When I tried to assign it as tensors, it gave me this error:

TypeError: Cannot convert value <tf.Tensor 'Const:0' shape=(1, 128) dtype=int32> to a TensorFlow DType.