XGBoost 0.7 release, Travis Oliphant’s new startup, Deepmind’s dm_control, Community connectors and more in today’s top stories around machine learning, deep learning, and data science news.
DeepMind released dm_control, the Deepmind Control Suit designed as performance benchmarks for reinforcement learning agents. The DeepMind Control Suite is a set of python reinforcement learning environments powered by the MuJoCo physics engine. It also includes libraries that provide Python bindings to the MuJoCo physics engine. The continuous control tasks are used for the design and performance comparison of reinforcement learning algorithms and are easy to use and modify. It includes a standardized structure and interpretable rewards, serving as performance benchmarks for RL agents. The uniform reward structure allows for robust suite-wide performance measures.
PostgreSQL was declared the DBMS of the year 2017 by DB-Engines, an information portal on database management systems. The DB-Engines Ranking is a monthly list of DBMS ranked by their current popularity. In 2017, PostgreSQL gained the most popularity from a pool of 341 monitored systems and was ranked number 1.
DB-Engines derive popularity scores for each DBMS by subtracting the popularity scores of the previous scores from the latest scores. These scores are calculated based on a number of parameters.
In 2017, PostgreSQL gained 55.81 scoring points outperforming all other systems in 2017. The new release of PostgreSQL 10 is considered to be a major factor behind this ranking. PostgreSQL 10 specifically focused on enhancements for effectively distributing data across many nodes.
A new feature rolled out with Google Data studio named as community connectors. Google Data studio allows users to build free live, interactive dashboards. Using this, users can fetch their data from multiple sources and create unlimited reports in data studio, with full editing and sharing capabilities. The addition of community connectors lets users explore the Apps Script to build connectors to any internet accessible data source. Users can also share the community connectors with others in order to access their own data from within Data Studio.
Some of the reasons why you should use Google’s Community Connectors are, to leverage Data Studio as a reporting platform for your customers, to reach a larger audience, and to develop customized enterprise solutions.
To know more about how to build your community connectors visit the link here.
Travis Oliphant, the Anaconda founder plans to unveil a new startup for helping organizations apply AI and ML to enterprise problems.
Travis Oliphant, the founder of Anaconda and creator of popular python libraries Numpy, Scipy and Numba, tweeted that he is leaving full-time employment in Anaconda as of 01-01-2018. Anaconda is an open Data Science platform powered by Python. Travis was responsible for directing Community Innovation across the company while providing oversight over the data science platform.
In 2018, he will be working on improving OSS sustainability through non-profit work and is starting a new services/product incubation company to help organizations make better use of OSS and AI/ML.
His new company, Quansight is a services firm that helps companies take advantage of Open Source Software (OSS) for obtaining quantitative insight on their data. They advise organizations on using OSS effectively (including all the libraries in Anaconda) and help them in applying the capabilities of artificial intelligence and modern machine learning to their biggest problems. They also connect domain experts to companies and ideas and incubate enterprise products.
XGBoost v0.7 is officially here!
XGBoost released the 0.7 version of their popular open-source gradient boosting framework. XGBoost is short for Extreme Gradient Boosting. The new 0.7 version represents a major change from the last release (v0.6), which was released 18 months ago.
The new features include
- Updated Sklearn API
- Refactored gbm to allow more friendly cache strategy
- Robust DMatrix construction from a sparse matrix
- Faster construction of DMatrix from 2D NumPy matrices: elide copies, use of multiple threads
- Automatic removal of nan from input data when it is sparse.
- Fixing of single-instance prediction function to obtain correct predictions
- Factoring out of Predictor interface (in a manner similar to the updater interface).
- Makefile support for Solaris and ARM
- Test code coverage using Codecov
- CPP tests added
- Dockerfile and Jenkinsfile to support continuous integration for GPU code
- New parameters in Python and R package
- Updated Documentation
More details about the new version update can be found at their github.
The Numba version 0.36.2 patch release was announced by Stanley Seibert. This release has only two minor patches. Firstly, it adds support for the CUDA 9.x toolkit, and secondly, it fixes the syntax error with the exec causing import errors in Python 2.7.0 through 2.7.8.
Numba will now work on the new toolkit–if present, though the community has not still released the cudatoolkit 9 conda packages into the Anaconda default channel, as it is still under some testing.
There are also some CUDA compatibility changes to look at. The community will drop the support for CUDA 7.5 and older in Numba once it officially starts supporting CUDA 9 in Anaconda. Also, NVIDIA has dropped support for compute capability 2.x from CUDA 9, so Numba will not work with these older cards when the CUDA 9 toolkit is present. Note that newer NVIDIA GPU drivers are backward compatible with older toolkits.
One cannot use the CUDA 9.1 toolkit unless the GPU drivers are new enough. Since GPU drivers can only be upgraded by system administrators, this is why the community supports a range of CUDA toolkits in Anaconda (and Numba).
Atonix Digital focuses on software development, sales, and delivery, as well as innovative analytics through the proprietary ASSET360® data analytics platform developed by Black & Veatch.
The cloud-based ASSET360® captures integrate and analyzes data to generate actionable information while providing comprehensive system awareness to businesses and municipalities, boosting operational performance.
Atonix’s software solutions aim at removing guesswork by easily modeling future scenarios and predicting outcomes often within seconds. It provides assistance for sharp planning, risk analysis, and budget control. Atonix also gives businesses a data-driven edge that brings into focus trends, issues and opportunities while maximizing efficiency, revenue, and performance.