Tensorflow 1.5.0, DataNucleus AccessPlatform 5.1.6, Databricks comes to Microsoft Azure, Citus 7.2, and more in today’s top stories around machine learning, deep learning, and data science news.
1. Tensorflow 1.5.0 now generally available with preview of Tensorflow Lite
Tensorflow 1.5.0 is now generally available. Previously, Tensorflow 1.5 RC was announced on 4th January, 2018. With Tensorflow 1.5, a lot of new features and changes have been added. The breaking changes revolve around prebuilt binaries. They are now built against CUDA 9 and cuDNN 7. Also, starting from 1.6 release, prebuilt binaries will use AVX instructions. Other major features and improvements include:
- The availability of eager execution preview version.
- The availability of TensorFlow Lite dev preview.
- Accelerated Linear Algebra (XLA) related changes.
- Addition of streaming_precision_recall_at_equal_thresholds, a method for computing streaming precision and recall with O(num_thresholds + size of predictions) time and space complexity.
- RunConfig default behavior will now not set a random seed, making random behavior independently random on distributed workers.
- The implementation of tf.flags is replaced with absl.flags.
- Support for CUBLAS_TENSOR_OP_MATH in fp16 GEMM.
- Support for CUDA on NVIDIA Tegra devices.
For details on bug fixes and other changes, see the full release notes here.
2. A new version of DataNucleus AccessPlatform is now available
DataNucleus AccessPlatform is Apache 2 licensed and provides retrieval of Java objects to a range of datastores using JDO/JPA/REST APIs, with a range of query languages. Now they have released the version 5.1.6 which includes a lot of new enhancements and bug fixes.
- ClassUtils.getConstructorWithArguments doesn’t allow to skip type check of one of the arguments.
- Support for queries with “IS NULL” / “IS NOT NULL” added.
- Support for String toUpperCase/toLowerCase/trim/trimLeft/trimRight/substring in JDOQL/JPQL added.
- Support for Numeric cos/sin/tan/acos/asin/atan/toDegrees/toRadians in JDOQL/JPQL added.
- Unable to execute an UPDATE JPQL Query against a domain class that contains ‘Set’ in its name.
- Lists might appear empty while they are actually not (forEach).
- Retrieval code doesn’t handle primitive retrieval when not existing in database
- Inequality Filter method, .ne() gives QueryExecutionException.
- Query with candidate being base of inheritance tree using “complete-table” strategy fails when overriding the “id” column name.
- JDOQL query fails when using reference to interface field, and implementations share table.
- @Basic @Lob ArrayList<byte[]> entity field results in erroneous metamodel.
- @Basic @Lob Serializable entity field results in erroneous metamodel.
The entire changelog can be found in their release notes.
3. Databricks integrates with Microsoft Azure
Until now, services of Databricks were available as a single cloud offering based on the Amazon Web Services (AWS) cloud. Starting 27th January 2018, a new flavor of Apache Spark service is announced. Called Azure Databricks (ADB), it is based on and is tightly integrated with Microsoft Azure. The Apache Spark-based analytics platform is optimized for the Microsoft Azure cloud services platform. It provides one-click setup, streamlined workflows, and an interactive workspace allowing data scientists, data engineers, and business analysts to collaborate. This new service is a first-party offering from Microsoft. It consists of three major parts, a notebook-based collaborative workspace, the Databricks Runtime, and a serverless compute model. ADB has direct support for Azure Blob Storage and Azure Data Lake Store. It also integrates with Cosmos DB and Azure Active Directory. More information is available here.
4. A new version of Citus (7.2), the distributed database is now released
Citus have announced the version 7.2 of their distributed database. With Citus database version 7.2, distributed SQL support is added to queries that run on data spread across a cluster of machines. A quick overview of the changes in Citus database version 7.2 for distributed queries include:
- Common Table Expressions (CTEs).
- Complex subqueries.
- Set operations (UNION, INTERSECT, etc).
- Joins between distributed and local tables through CTEs.
- Joins that include non-equality clauses.
- Partition management automation with pg_partman.
Citus 7.2 is compatible with PostgreSQL 9.6 and 10. It can be downloaded by following the instructions here. It can also be deployed in a single-click through the Citus Cloud console. To learn more about the new features, visit the official blog.
5. CapLinked announces Transitnet, a new blockchain framework
CapLinked has launched their new blockchain framework TransitNet to protect and record enterprise transactions. TransitNet is a decentralized protocol that protects digital assets and permanently records data access. The protocol is accessible via an API and is used to apply protections and activity tracking for information exchanging during business deals. TransitNet adds security to transfer of funds, as opposed to Ripple and other decentralized technologies that are addressing payments. TransitNet’s decentralized application will allow users to apply protections and track their digital assets when they’re transferred to third parties. They’ll be able to encrypt, watermark, and set access parameters for digital assets being moved and track their movement on an immutable decentralized ledger.