tensor board

17
Visualizing your Deep learning using TensorBoard (TensorFlow) Sung Kim <[email protected]>

Upload: sung-kim

Post on 21-Apr-2017

4.428 views

Category:

Engineering


2 download

TRANSCRIPT

Page 1: Tensor board

Visualizing your Deep learning using TensorBoard (TensorFlow)

Sung Kim <[email protected]>

Page 2: Tensor board

TensorBoard: TF logging/debugging tool

• Visualize your TF graph

• Plot quantitative metrics

• Show additional data

https://www.tensorflow.org/versions/r0.7/how_tos/summaries_and_tensorboard/index.html

Page 3: Tensor board

Old fashion: print, print, print

Page 4: Tensor board

New way!

Page 5: Tensor board

5 steps of using tensorboard

• From TF graph, decide which node you want to annotate- with tf.name_scope("test") as scope:

- tf.histogram_summary("weights", W), tf.scalar_summary(“accuracy", accuracy)

• Merge all summaries- merged = tf.merge_all_summaries()

• Create writer- writer = tf.train.SummaryWriter("/tmp/mnist_logs", sess.graph_def)

• Run summary merge and add_summary- summary = sess.run(merged, …); writer.add_summary(summary);

• Launch Tensorboard- tensorboard --logdir=/tmp/mnist_logs

Page 6: Tensor board

Name variables

Page 7: Tensor board

Add scope for better graph hierarch

Page 8: Tensor board

Add histogram

Page 9: Tensor board

Add scalar variables

Page 10: Tensor board

Merge summaries and create writer after creating session

Page 11: Tensor board

Run merged summary and write (add summary)

Page 12: Tensor board

Launch tensorboard

• tensorboard —logdir=/tmp/mnist_logs

• (You can navigate to http://0.0.0.0:6006)

Page 13: Tensor board

Add scope for better graph hierarch

Page 14: Tensor board

Add scope for better graph hierarch

Page 15: Tensor board

Add histogram

Page 16: Tensor board

Add scalar variables

Page 17: Tensor board

5 steps of using tensorboard

• From TF graph, decide which node you want to annotate- with tf.name_scope("test") as scope:

- tf.histogram_summary("weights", W), tf.scalar_summary(“accuracy", accuracy)

• Merge all summaries- merged = tf.merge_all_summaries()

• Create writer- writer = tf.train.SummaryWriter("/tmp/mnist_logs", sess.graph_def)

• Run summary merge and add_summary- summary = sess.run(merged, …); writer.add_summary(summary);

• Launch Tensorboard- tensorboard --logdir=/tmp/mnist_logs