Seldon

This release provides the ability to create predictive pipelines for multi-class classification models.
- Create feature extraction and manipulation pipelines in python to create appropriate features for training machine learning models. Automatically load and run the same transformations at runtime when receiving features to provide predictions on. Feature transformations include:
- TFIDF feaures with chi-squared feature selection
- Automatic detection of categorical, date and numeric features with normalisation of numeric features
- Simple pipeline and transformation classes that can be extended to create custom feature transformations
- Create classification models using Vowpal Wabbit and XGBoost
- Example microservices for runtime scoring that load and run feature pipelines and predict against Vowpal Wabbit and XGBoost models

For further technical docs please see: http://docs.seldon.io/prediction-overview.html
We provide a demo for creating a multi-class classification predictive endpoint for the classic Iris classification task: http://docs.seldon.io/iris-demo.html


Bug fix release

Optimization minor release.
- less default logging
- configurable number of spymemcached clients
- central ModelManager to load models

This release provides a new Association Rule Recommender useful for e-commerce settings.
See http://docs.seldon.io/spark-models.htmlassoc-rules

- tag affinity based recommender
- performance optimization updates
- multiple item dimension handling

General Prediction Endpoint

Until now, Seldon has been focused on providing an enterprise-grade open-source recommendation engine – i.e. to suggest articles, videos, products to people based on behavioural and contextual data.

In version 0.93, a general prediction endpoint is now available to developers. This major new feature allows easy integration of classification and regression machine learning models into the Seldon platform for runtime scoring. In this initial release, we provide the ability to load and score [Vowpal Wabbit classification models inside the Seldon server](http://docs.seldon.io/runtime-prediction.html).

You can update models in production with no downtime. We have also created a simple microservice REST API to make it straightforward to integrate existing machine learning toolkits. As an example, we show [how to integrate Vowpal Wabbit running a model in daemon mode](http://docs.seldon.io/pluggable-prediction-algorithms.htmlprediction-python-vw).

We plan to provide further examples of integrating toolkits via the [microservice REST API](http://docs.seldon.io/pluggable-prediction-algorithms.html) as well as further extend the Seldon server itself to load and store a range of popular models.

Seldon AWS AMI Private Access Program

Our Seldon AWS AMI private access program continues to grow as more users choose to get up and running quickly using the AMI. To participate, [register for access](http://docs.seldon.io/vm-aws.html).

Users that have registered and received access to the Seldon AWS AMI will get access enabled to new AMI releases when they become available. Check the Seldon user group for details of the latest launch URLs.

Seldon Virtual Machine 0.93 release

We are pleased to announce an exciting new version of the Seldon virtual machine. This release makes the Seldon Platform more accessible and easier to get up and running.
There are two flavours to choose from:
1. On the desktop there is [a Vagrant install](http://docs.seldon.io/vm.html).
2. On Amazon Web Services there [a pre-configured AMI](http://docs.seldon.io/vm-aws.html).

Both these virtual machines now use Ubuntu and allows developers to extend and customize the system as necessary.

The size of the Vagrant download has now been significantly reduced by utilizing Docker Hub for additional content.

We look forward to hearing from developers who are working with this functionality. If you have any questions or feedback, please send them to the [Seldon user group](http://bit.ly/seldon-users) or email [devseldon.io](mailto:devseldon.io).