Demos illustrating ml4j-quickstart functionality... more demos to be added soon.
For showcases of more complex configurations, such as the InceptionV4 network, or the YOLOv2 network, please see:
https://github.com/ml4j/inception-v4-spring-demo
https://github.com/ml4j/yolo-v2-spring-demo
Download the jar though Maven:
<repository>
<id>ml4j-snapshots</id>
<url>https://raw.githubusercontent.com/ml4j/mvn-repository/master/snapshots</url>
<snapshots>
<enabled>true</enabled>
</snapshots>
</repository><dependency>
<groupId>org.ml4j</groupId>
<artifactId>ml4j-quickstart</artifactId>
<version>2.0.0-SNAPSHOT</version>
</dependency>The ml4j-api-2.0.0.RC1 release candidate has now been released, and there are to be no planned changes to the API contract of any of the ml4j-api components before the ml4j-api-2.0.0 final release. Most of the changes left to be made to the api are for Javadoc and unit tests.
The implementations in the various impl projects are still in snapshot status (eg. ml4j-impl, ml4j-default-components ). These projects are almost fully functional, with a couple of exceptions - for example, while batch norm is available for graph networks (eg. Inception networks), it has not yet been implemented for sequential layer networks - also ResidualBlocks have not yet been implemented).
This remaining to-do functionality should be a matter of re-using the already written components from other parts of the project - however the recent focus has been on delivering a stable API contract in the ml4j-api-2.0.0.RC1 release.
The main technical debt with the impl projects is with unit testing and Javadoc, the coverage of which will need to be much higher before a stable impl release can be delivered.
The various demo projects have acted as integration tests throughout development - eg. the demos in following projects:
https://github.com/ml4j/inception-v4-spring-demo https://github.com/ml4j/yolo-v2-spring-demo https://github.com/ml4j/ml4j-neuralnets-demo