TensorboardX Versions Save

Remove a dependency issue caused by v2.3. (#631)

2.3 (2021-06-20)

• Support logging to comet.ml simutaneously.

• Support for type hints.
• Dropped Python 2 support.
• Bug fixes, see the commit log in Github.

2.0 (2019-12-31)

• Now you can tag Hparams trials with custom name instead of the default epoch time
• Fixed a bug that add_hparams are rendered incorrectly with non-string values
• Supports logging to Amazon S3 or Google Cloud Storage
• Bug fixes and error message for add_embedding function
• Draw openvino format with add_openvino_graph

2.1 (2020-07-05)

• Global SummaryWriter that mimics python's default logger class, concurrent write is supported.
• 200x speed up for add_audio. Please install the soundfile package for this feature.
• Supports jax tensors.
• The add_graph function is delegated to the one in torch.utils.tensorboard.
• Bug fixes, see the commit log in Github.

• Use new JIT backend for pytorch. This works better with pytorch 1.2 and 1.3
• Supports hparams plugin
• add_embedding now supports numpy array input

1.8 (2019-07-05)

• Draw label text on image with bounding box provided.
• crc32c speed up (optional by installing crc32c manually)
• Rewrite add_graph. onnx backend is replaced by JIT to support more advanced structure.
• Now you can add_mesh() to visualize colorful point cloud or meshes.

• Able to write to S3
• Fixed raw histogram issue that nothing is shown in TensorBoard
• Users can use various image/video dimension permutation by passing 'dataformats' parameter.
• You can bybass the writer by passing write_to_disk=True to SummaryWriter.
• Fix bugs

[update Jun.21 2019] important notice

SummaryWriter's initializer's paremeter: log_dir was accidently changed to logdir. This makes some code not backward compatible. The unit tests doesn't cover that so It just got merged and be part of v1.7.

• Many graph related bug is fixed in this version.
• New function: add_images(). This function accepts 4D iamge tensor. See documentation.
• Make add_image_with_boxes() usable.
• API change: add_video now accepts BxTxCxHxW instead of BxCxTxHxW tensor.