Using TensorFlow TensorBoard

Note: The purpose of this post is as a personal reflection and not as a tutorial.

When doing machine learning research, it is important to be able to have real time visualisation. TensorBoard allows such visualisation. This post will document my experiments utilising TensorBoard.

TensorBoard Guides

I Looked at these guides:

• TensorBoard Getting Started Guide
• TensorBoard Scalars: Logging training metrics in Keras
• Displaying image data in TensorBoard
• Examining the TensorBoard Graph

These TensorFlow API references are also relevant:

• Writing to logs: tf.summary
• Understannding callbacks in keras: tf.keras.callbacks

Logging Audio

TensorFlow API reference for logging audio.

• speech_4 was a list of 1D numpy arrays representing mono audio.
• transcripts was a list of strings representing the spoken characters in the corresponding audio.

# Clear out any prior log data.
!rm -rf logs

# Sets up a timestamped log directory.
logdir = "logs/train_data/" + datetime.now().strftime("%Y%m%d-%H%M%S")
# Creates a file writer for the log directory.
file_writer = tf.summary.create_file_writer(logdir)

for i in range(4):
  print(i)
  audio = speech_4[i]
  audio = audio.reshape((1, -1, 1))
  description = '**Transcript**: {}\n\n'.format(transcripts[i])
  with file_writer.as_default():
    tf.summary.audio("Audio " + str(i+1), audio, 48000, step=0, description=description)

How the audio appears in TensorBoard:

• The description with the transcript can be viewed by clicking the information symbols.