How to log a tensorflow layer output in tf.estimator.Estimator()

Question

In this tutorial, they successfully log the softmax function by giving a name to the tf.nn.softmax node.

tf.nn.softmax(logits, name="softmax_tensor") # giving name to the node
.
.
.
tensors_to_log = {"probadfabilities": "softmax_tensor"} # logging the node

logging_hook = tf.train.LoggingTensorHook(
    tensors=tensors_to_log, every_n_iter=50)
eval_input_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": eval_data},
    y=eval_labels,
    num_epochs=1,
    shuffle=False)

Now, instead of the softmax, I would like to also log the output of the last Dense layer.

logits = tf.layers.dense(inputs=dropout, units=10, name='logits')
.
.
.
tensors_to_log = {"last_layer": "logits"}

But it gives me the following error

KeyError: "The name 'logits:0' refers to a Tensor which does not exist. The operation, 'logits', does not exist in the graph."

My question is: how to log the layer output in tensorflow?

My code

import tensorflow as tf
import numpy as np
import os

tf.logging.set_verbosity(tf.logging.INFO)

def cnn_model_fn(features, labels, mode):
    """Model function for CNN."""
    # Input Layer
    input_layer = tf.reshape(features["x"], [-1, 28, 28, 1])

    # Convolutional Layer #1
    conv1 = tf.layers.conv2d(
      inputs=input_layer,
      filters=128,
      kernel_size=[7, 7],
      padding="same",
      activation=tf.nn.relu)

    # Pooling Layer #1
    pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2)

    # Convolutional Layer #2 and Pooling Layer #2
    conv2 = tf.layers.conv2d(
      inputs=pool1,
      filters=256,
      kernel_size=[5, 5],
      padding="same",
      activation=tf.nn.relu)
    pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2)

    # Dense Layer
    pool2_flat = tf.reshape(pool2, [-1, 7 * 7 * 256])
    dense = tf.layers.dense(inputs=pool2_flat, units=1024, activation=tf.nn.relu)
    dropout = tf.layers.dropout(
      inputs=dense, rate=0.4, training=mode == tf.estimator.ModeKeys.TRAIN)

    # Logits Layer
    logits = tf.layers.dense(inputs=dropout, units=10, name='logits')

    if mode == tf.estimator.ModeKeys.PREDICT:
        return tf.estimator.EstimatorSpec(mode=mode, predictions=predictions)

    # Calculate Loss (for both TRAIN and EVAL modes)
    loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)

    # Configure the Training Op (for TRAIN mode)
    if mode == tf.estimator.ModeKeys.TRAIN:
        optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.001)
        train_op = optimizer.minimize(
            loss=loss,
            global_step=tf.train.get_global_step())
        return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)

    # Add evaluation metrics (for EVAL mode)
    eval_metric_ops = {
      "accuracy": tf.metrics.accuracy(
          labels=labels, predictions=predictions["classes"])
    }
    return tf.estimator.EstimatorSpec(
      mode=mode, loss=loss, eval_metric_ops=eval_metric_ops)

((train_data, train_labels),
 (eval_data, eval_labels)) = tf.keras.datasets.mnist.load_data()

train_data = train_data/np.float32(255)
train_labels = train_labels.astype(np.int32)  # not required

eval_data = eval_data/np.float32(255)
eval_labels = eval_labels.astype(np.int32)

mnist_classifier = tf.estimator.Estimator(
    model_fn=cnn_model_fn, model_dir="./mnist_convnet_model")

# Set up logging for predictions
tensors_to_log = {"last_layer": "logits"}

logging_hook = tf.train.LoggingTensorHook(
    tensors=tensors_to_log, every_n_iter=50)

# Train the model
train_input_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": train_data},
    y=train_labels,
    batch_size=100,
    num_epochs=None,
    shuffle=True)

# train one step and display the probabilties
mnist_classifier.train(
    input_fn=train_input_fn,
    steps=10,
    hooks=[logging_hook])

Answer 1

In the tf.official ResNet implementation, they use tf.identity for this purpose:

logits = tf.identity(logits, 'logits')