AWS SageMaker Frameworks and Algorithms Refresh

Amazon Web Services (AWS), announced last week that it is making several improvements to SageMaker, its machine learning modeling platform.
These improvements are made to the program’s BlazingText and DeepAR algorithms as well as Linear Learning algorithms.
AWS stated that it has improved the accuracy and ease of DeepAR’s algorithms for forecasting, so that “missing value are now handled within this model.” DeepAR now supports seasonality patterns, custom time-varying features, and multiple groups of time series.
BlazingText has been enhanced with the addition of a Word2Vec algorithm, which optimizes the use of GPU hardware, according to the company.
Other improvements include “meaningful Vectors for Out-of-vocabulary Words (OOV) that are not in the Training Dataset” and new support to “high-speed multiclass and multi-label text classification,” among other improvements.
Linear Learning users will now be able to use multi-class classifications and other improved methods of sorting and classifying their data.
SageMaker now supports Chainer 4.1, offering pre-configured containers in which users will find Layerwise Adaptive Rat Scaling (LARS). AWS claims that this improves the training of networks with large batch sizes.
These changes are currently being implemented across all AWS instances around the world.