Keras 2.1.4, Updates to TensorFlow Object detection API, PyTorch 0.3.1, New releases in Pgpool-II, and more in today’s top stories around machine learning, deep learning, and data science news.
1. Keras 2.1.4 releases
Keras 2.1.4 is released with bug fixes and improvements in performance and example scripts. The major changes include:
- In ImageDataGenerator, change default interpolation of image transforms from nearest to bilinear.
- Stateful metrics in model.compile(…, metrics=[…]) are now allowed. A stateful metric inherits from Layer, and implements __call__ and reset_states.
- Support for constants argument in StackedRNNCells.
- Some TensorBoard features are enabled in the TensorBoard callback (loss and metrics plotting) with non-TensorFlow backends.
- Reshape argument in model.load_weights() are added, to optionally reshape weights being loaded to the size of the target weights in the model considered.
The entire changes are available in the release notes.
2. TensorFlow Object Detection API gets updated with instance segmentation
Tensorflow announced the addition of instance segmentation to their object detection API. Instance segmentation is used to segment an object region once it is detected. Instance segmentation allows for more fine-grained information about the extent of the object within the box. With this API update, Tensorflow now supports a number of instance segmentation models similar to those discussed in the Mask R-CNN paper. The model now predicts masks in addition to object bounding boxes. They have also provided four instance segmentation config files to be used to train models:
More details can be read at the official Github repo.
3. PyTorch 0.3.1 release
PyTorch have released a minor release 0.3.1 of bug fixes and performance improvements.
- They have removed support for CUDA capability 3.0 and 5.0
- Binary releases for CUDA 7.5 are now stopped.
- They will now add CPU-only binary releases that are 10x smaller in size than the full binary with CUDA capabilities.
- Added Cosine Annealing Learning Rate Scheduler
- Added reduce argument to PoissonNLLLoss to be able to compute unreduced losses
- Added random_split that randomly splits a dataset into non-overlapping new datasets of given lengths
- Introduced scopes to annotate ONNX graphs to have better TensorBoard visualization of models
- Allowed map_location in torch.load to be a string, such as map_location=’cpu’ or map_location=’cuda:2′
Bug fixes and other improvements are available in the changelog.
4. Pgpool-II 3.7.2, 3.6.9, 3.5.13, 3.4.16 and 3.3.20 are now officially released
Pgpool-II is a tool to add useful features to PostgreSQL, such as connection pooling, load balancing, and automatic failover. Pgpool Global Development Group has announced the availability of versions 3.7.2, 3.6.9, 3.5.13, 3.4.16, and 3.3.20 of Pgpool-II. The changes include:
- Fixed the bug with socket writing added in Pgpool-II 3.7.0, 3.6.6 and 3.5.10.
- Allow building with libressl.
- Set TCP_NODELAY and non-blocking to frontend socket.
- TCP_NODELAY is now employed
- Changed systemd service file to use STOP_OPTS=” -m fast”.
- Changed pgpool_setup to add restore_command in recovery.conf.
For more information, take a look at the release notes.
5. OmniDB 2.5 is now released with support for Oracle Databases
OmniDB 2.5, the browser-based database management tool, is now released. It now allows users to manage multiple databases in a unified workspace with a user-friendly and fast-performing interface.
The following features and improvements are added:
- Basic support for Oracle databases. Users can manage, connect, and interact with Oracle databases using most of the same features provided to manage PostgreSQL databases.
- New DDL Panel. A new panel located below the treeview displays properties and DDL of the currently selected node.
For a complete list of updates, read the OmniDB change tracker.