Logging to tensorboard without tensorflow operations. Uses manually generated summaries instead of summary ops

tensorboard_logging.py

"""Simple example on how to log scalars and images to tensorboard without tensor ops.

License: BSD License 2.0
"""
__author__ = "Michael Gygli"

import tensorflow as tf
from StringIO import StringIO
import matplotlib.pyplot as plt
import numpy as np

class Logger(object):
    """Logging in tensorboard without tensorflow ops."""

    def __init__(self, log_dir):
        """Creates a summary writer logging to log_dir."""
        self.writer = tf.summary.FileWriter(log_dir)

    def log_scalar(self, tag, value, step):
        """Log a scalar variable.

        Parameter
        ----------
        tag : basestring
            Name of the scalar
        value
        step : int
            training iteration
        """
        summary = tf.Summary(value=[tf.Summary.Value(tag=tag,
                                                     simple_value=value)])
        self.writer.add_summary(summary, step)

    def log_images(self, tag, images, step):
        """Logs a list of images."""

        im_summaries = []
        for nr, img in enumerate(images):
            # Write the image to a string
            s = StringIO()
            plt.imsave(s, img, format='png')

            # Create an Image object
            img_sum = tf.Summary.Image(encoded_image_string=s.getvalue(),
                                       height=img.shape[0],
                                       width=img.shape[1])
            # Create a Summary value
            im_summaries.append(tf.Summary.Value(tag='%s/%d' % (tag, nr),
                                                 image=img_sum))

        # Create and write Summary
        summary = tf.Summary(value=im_summaries)
        self.writer.add_summary(summary, step)
        

    def log_histogram(self, tag, values, step, bins=1000):
        """Logs the histogram of a list/vector of values."""
        # Convert to a numpy array
        values = np.array(values)
        
        # Create histogram using numpy        
        counts, bin_edges = np.histogram(values, bins=bins)

        # Fill fields of histogram proto
        hist = tf.HistogramProto()
        hist.min = float(np.min(values))
        hist.max = float(np.max(values))
        hist.num = int(np.prod(values.shape))
        hist.sum = float(np.sum(values))
        hist.sum_squares = float(np.sum(values**2))

        # Requires equal number as bins, where the first goes from -DBL_MAX to bin_edges[1]
        # See https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/framework/summary.proto#L30
        # Thus, we drop the start of the first bin
        bin_edges = bin_edges[1:]

        # Add bin edges and counts
        for edge in bin_edges:
            hist.bucket_limit.append(edge)
        for c in counts:
            hist.bucket.append(c)

        # Create and write Summary
        summary = tf.Summary(value=[tf.Summary.Value(tag=tag, histo=hist)])
        self.writer.add_summary(summary, step)
        self.writer.flush()

Example usage:

    import numpy as np
    logger = Logger('/tmp/test')
    for i in range(1000):
        l.log_histogram('test_hist',np.random.rand(50)*(i+1),i)

Then simpy run tensorboard:

    tensorboard --logdir /tmp/test

Gives the following histogram on tensorboard (and the corresponding distribution)

This is great!

Any chance you can add a function for text?

Could you include a license?

Thanks or the code it works great, found a small bug though...
If values is a list in log_histogram then because list has no attribute shape the program crashes you should write values=np.array(values) in the beginning of the function to avoid these issue.

@nasimrahaman: Added copyleft
@MiguelMonteiro fixed

Thank you for such a helpful tool ! I've been confused about how to write what I want into tensorboard

Thanks!

            s = StringIO()
            plt.imsave(s, img, format='png')

doesn't work because pyplot.imsave() expects a file path

doesn't work because pyplot.imsave() expects a file path

That's not true, at least for more recent versions, where it can be a "file-like" object, see:
https://matplotlib.org/api/_as_gen/matplotlib.pyplot.imsave.html

@adizhol: Maybe your version is outdated?

Thanks!

It works, thx!

Great code, thanks

excellent work, and I have to use BytesIO instead of StringIO for python 3.5.5, matplotlib 2.2.2

Cool! Thanks

@gyglim Maybe it is worth to add flush method? I think for newcomers to tensorboard it would be a good hint that they should consider flushing.

Thanks a lot, perfect!

Thanks a lot Michael, this helped me today :).
BTW: I had to use a Bytes-like object (BytesIO) instead of a StringIO for it to work (TF 1.13, Matplotlib 3.0.3).

Cheers !

Are you sure this works for multiple scalars? Is flush needed? I see points showing up in tboard but they do not seem to update often enough?

Also anyone know what to put in a step to get the scalars to look similar to the keras callback scalars?

text summary can be a new feature

Anyone knows how to construct text summary?

What does 'License: Copyleft' mean?
Can you provide more detailed license information, such as GPL, LGPL or MPL?

Thanks!!

Added BSD License 2.0

@gyglim Thanks!! :)

To add text summary

from tensorboard.plugins.text import metadata

class Logger(object):
    """Logging in tensorboard without tensorflow ops."""

    def __init__(self, log_dir):
        """Creates a summary writer logging to log_dir."""
        self.writer = tf.summary.FileWriter(log_dir)

	def log_text(self, tag, value, step):
	        """Log string or 2D string tables. """
	        summary_metadata = metadata.create_summary_metadata(
	                display_name="text",
	                description="Text Summary")
	        summary_metadata = tf.SummaryMetadata.FromString(
	                summary_metadata.SerializeToString())
	        tensor = tf.make_tensor_proto(value, dtype=tf.string)
	        summary = tf.Summary(value=[tf.Summary.Value(
	            tag=tag,
	            metadata=summary_metadata,
	            tensor=tensor
	            )])
	        self.writer.add_summary(summary, step)
	        self.writer.flush()

Example usage:

logger = Logger('/tmp/test')
logger.log_text(tag="string", value="test", step=0)
logger.log_text(tag="table", value=[["r0c0", "r0c1"], ["r1c0", "r1c1"]], step=0)

Tensorboard: