Tensorboard

TensorBoard is a visualization tool provided by Google TensorFlow, designed to help understand, debug, and optimize machine learning models. It displays various metrics and data during the training process through a graphical interface, allowing developers to intuitively understand the performance and behavior of their models.

You can synchronize projects tracked with Tensorboard to SwanLab in two ways:

• Synchronous Tracking: If your current project uses Tensorboard for experiment tracking, you can use the swanlab.sync_tensorboardX() or swanlab.sync_tensorboard_torch() commands to synchronize metrics to SwanLab while running your training script.
• Convert Existing Projects: If you want to copy a project from Tensorboard to SwanLab, you can use swanlab convert to convert a directory containing TFevent files into a SwanLab project.

INFO

The current version only supports converting scalar and image charts.

• 1. Synchronous Tracking
  • 1.1 TensorboardX: Add the sync_tensorboardX Command
  • 1.2 PyTorch: Add the sync_tensorboard_torch Command
  • 1.3 Alternative Approach
  • 1.4 Test Code
• 2. Convert Existing Projects
  • 2.1 Method 1: Command Line Conversion
  • 2.2 Method 2: In-Code Conversion
  • 2.3 Parameter List

1. Synchronous Tracking

1.1 TensorboardX: Add the sync_tensorboardX Command

If you are using TensorboardX, you can add the swanlab.sync_tensorboardX() command anywhere before executing tensorboardX.SummaryWriter() to synchronize metrics to SwanLab during training.

import swanlab
from tensorboardX import SummaryWriter

swanlab.sync_tensorboardX()

writer = SummaryWriter(log_dir='./runs')

1.2 PyTorch: Add the sync_tensorboard_torch Command

If you are using PyTorch's built-in Tensorboard, you can add the swanlab.sync_tensorboard_torch() command anywhere before executing torch.utils.tensorboard.SummaryWriter() to synchronize metrics to SwanLab during training.

import swanlab
import torch

swanlab.sync_tensorboard_torch()

writer = torch.utils.tensorboard.SummaryWriter(log_dir='./runs')

1.3 Alternative Approach

You can also manually initialize SwanLab first and then run the Tensorboard code.

TensorboardX

import swanlab
from tensorboardX import SummaryWriter

swanlab.init(...)
swanlab.sync_tensorboardX()

...

writer = SummaryWriter(log_dir='./runs')

PyTorch

import swanlab
from torch.utils.tensorboard import SummaryWriter

swanlab.init(...)
swanlab.sync_tensorboard_torch()

...

writer = SummaryWriter(log_dir='./runs')

1.4 Test Code

TensorboardX

import swanlab
from tensorboardX import SummaryWriter

swanlab.sync_tensorboardX()

writer = SummaryWriter(log_dir='./runs')

epochs = 10
offset = random.random() / 5
for epoch in range(2, epochs):
  acc = 1 - 2 ** -epoch - random.random() / epoch - offset
  loss = 2 ** -epoch + random.random() / epoch + offset

  writer.add_scalar("acc", acc, epoch)
  writer.add_scalar("loss", loss, epoch)

PyTorch

import swanlab
from torch.utils.tensorboard import SummaryWriter

swanlab.sync_tensorboard_torch()

writer = SummaryWriter(log_dir='./runs')

epochs = 10
offset = random.random() / 5
for epoch in range(2, epochs):
  acc = 1 - 2 ** -epoch - random.random() / epoch - offset
  loss = 2 ** -epoch + random.random() / epoch + offset

  writer.add_scalar("acc", acc, epoch)
  writer.add_scalar("loss", loss, epoch)

2. Convert Existing Projects

2.1 Method 1: Command Line Conversion

swanlab convert -t tensorboard --tb_logdir [TFEVENT_LOGDIR]

Here, [TFEVENT_LOGDIR] refers to the path of the log files generated when you previously recorded experiments with Tensorboard.

The SwanLab Converter will automatically detect tfevent files in the specified directory and its subdirectories (default depth is 3) and create a SwanLab experiment for each tfevent file.

2.2 Method 2: In-Code Conversion

from swanlab.converter import TFBConverter

tfb_converter = TFBConverter(convert_dir="[TFEVENT_LOGDIR]")
tfb_converter.run()

This has the same effect as the command line conversion.

2.3 Parameter List

Parameter	Corresponding CLI Parameter	Description
convert_dir	-	Path to Tfevent files
project	-p, --project	SwanLab project name
workspace	-w, --workspace	SwanLab workspace name
cloud	--cloud	Whether to use the cloud version, default is True
logdir	-l, --logdir	Path to save SwanLab log files

Example:

from swanlab.converter import TFBConverter

tfb_converter = TFBConverter(
    convert_dir="./runs",
    project="Tensorboard-Converter",
    workspace="SwanLab",
    logdir="./logs",
    )
tfb_converter.run()

The equivalent CLI command:

swanlab convert -t tensorboard --tb_logdir ./runs -p Tensorboard-Converter -w SwanLab -l ./logs

Executing the above script will create a project named Tensorboard-Converter in the SwanLab workspace, convert the tfevent files in the ./runs directory into SwanLab experiments, and save the logs generated by SwanLab in the ./logs directory.