TensorBoard: Difference between revisions

From David's Wiki
No edit summary
 
(3 intermediate revisions by the same user not shown)
Line 1: Line 1:
TensorBoard is a way to visualize your model and various statistics during or after training.
TensorBoard is a way to visualize your model and various statistics during or after training.


==Custom Usage==
==CLI Usage==
===CLI===
<pre>
tensorboard --logdir [logs]
</pre>
 
;Flags
*<code>--samples_per_plugin</code> indices the number of samples to show for each tab. Non-scalar objects are sampled using reservoir sampling.
** <code>--samples_per_plugin images=10000</code> samples approximately 10000 images.
 
==Training Usage==
If you're using a custom training loop (i.e. gradient tape), then you'll need to set everything up manually.
If you're using a custom training loop (i.e. gradient tape), then you'll need to set everything up manually.


Line 15: Line 25:
with train_summary_writer.as_default():
with train_summary_writer.as_default():
   tf.summary.scalar("training_loss", m_loss.numpy(), step=int(ckpt.step))
   tf.summary.scalar("training_loss", m_loss.numpy(), step=int(ckpt.step))
</syntaxhighlight>
==PyTorch==
PyTorch also supports output tensorboard logs. 
See [https://pytorch.org/docs/stable/tensorboard.html https://pytorch.org/docs/stable/tensorboard.html]. 
There is also [https://github.com/lanpa/tensorboardX lanpa/tensorboardX] but I haven't tried it.
<syntaxhighlight lang="python">
from torch.utils.tensorboard import SummaryWriter
writer = SummaryWriter(log_dir="./runs")
writer.add_scalar("train_loss", loss_np, step)
# Optionally flush e.g. at checkpoints
writer.flush()
# Close the writer (will flush)
writer.close()
</syntaxhighlight>
</syntaxhighlight>



Latest revision as of 14:22, 11 November 2020

TensorBoard is a way to visualize your model and various statistics during or after training.

CLI Usage

CLI

tensorboard --logdir [logs]
Flags
  • --samples_per_plugin indices the number of samples to show for each tab. Non-scalar objects are sampled using reservoir sampling.
    • --samples_per_plugin images=10000 samples approximately 10000 images.

Training Usage

If you're using a custom training loop (i.e. gradient tape), then you'll need to set everything up manually.

First create a SummaryWriter

train_log_dir = os.path.join(args.checkpoint_dir, "logs", "train")
train_summary_writer = tf.summary.create_file_writer(train_log_dir)

Scalars

Add scalars using tf.summary.scalar:

with train_summary_writer.as_default():
  tf.summary.scalar("training_loss", m_loss.numpy(), step=int(ckpt.step))

PyTorch

PyTorch also supports output tensorboard logs.
See https://pytorch.org/docs/stable/tensorboard.html.
There is also lanpa/tensorboardX but I haven't tried it.

from torch.utils.tensorboard import SummaryWriter
writer = SummaryWriter(log_dir="./runs")

writer.add_scalar("train_loss", loss_np, step)

# Optionally flush e.g. at checkpoints
writer.flush()

# Close the writer (will flush)
writer.close()

Resources