"""%tensorboard line magic that patches TensorBoard's implementation to make use of Jupyter

TensorBoard server extension providing built-in proxying.

Use:

%load_ext tensorboard.notebook

%tensorboard --logdir /logs

"""

import argparse

import uuid

from IPython.display import display, HTML, Javascript

def _tensorboard_magic(line):

"""Line magic function.

Makes an AJAX call to the Jupyter TensorBoard server extension and outputs

an IFrame displaying the TensorBoard instance.

"""

parser = argparse.ArgumentParser()

parser.add_argument('--logdir', default='/kaggle/working')

args = parser.parse_args(line.split())

iframe_id = 'tensorboard-' + str(uuid.uuid4())

html = """

<!-- JUPYTER_TENSORBOARD_TEST_MARKER -->

<script>

const req = {

method: 'POST',

contentType: 'application/json',

body: JSON.stringify({ 'logdir': '%s' }),

headers: { 'Content-Type': 'application/json' }

};

const baseUrl = Jupyter.notebook.base_url;

fetch(baseUrl + 'api/tensorboard', req)

.then(res => res.json())

.then(res => {

const iframe = document.getElementById('%s');

iframe.src = baseUrl + 'tensorboard/' + res.name;

iframe.style.display = 'block';

});

</script>

<iframe

id="%s"

style="width: 100%%; height: 620px; display: none;"

frameBorder="0">

</iframe>

""" % (args.logdir, iframe_id, iframe_id)

display(HTML(html))

def load_ipython_extension(ipython):

"""IPython extension entry point."""

ipython.register_magic_function(

_tensorboard_magic,

magic_kind='line',

magic_name='tensorboard',

)