How Can I Integrate Tensorboard Visualization To Tf.Estimator?

I have classical TensorFlow code for recognizing handwritten digits https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/tutorials/mnist/mnist_with_summaries.py

Solution 1:

Generally, You just need to specify tf.summary.scalar(), tf.summary.histogram() or tf.summary.image() anywhere in the code. You can use histogram summary in the following way to capture all weights and biases

for value in tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES):
    tf.summary.histogram(value.name, value)

As for updatable metrics summary, e.g. accuracy of f1 score, you need to wrap it in eval_metric_ops and pass to tf.estimator.EstimatorSpec

accuracy = tf.metrics.accuracy(labels=labels, predictions=predictions)
    eval_metric_ops = {'accuracy': accuracy}

1. You can just call tensorboard with the same dir you specified during training.
2. You don't need to use tf.summary.merge_all()