This Java program showcases the following aspects of the Smile Machine Learning library:
- loading data from file
- transforming features/data frames
- fitting a regression model
- serializing/deserializing a model to/from binary representation using vanilla Java
- serializing a model to JSON using protostuff
- applying a trained regression model to predict values
- validating the model
This example relies on the Boston House Prices data set.
According to the docs, most Smile models implement the Java
Serializable interface. This allows them to be serialized to binary representation using vanilla Java, which is
demonstrated in this showcase.
Additionally, this implementation indicates how a Smile model can be serialized to JSON using protostuff. Protostuff seems to be the recommended way, as it is explicitly mentioned in the Smile docs in the context of (JSON) model serialization.
This showcase relies on native (C++) dependencies as Smile needs BLAS/LAPACK for its linear regression algorithms. Please refer to the pom.xml for details.
This showcase has been tested on Java 8 and Java 11. While Smile seems to run fine on both Java 8 and Java 11, protostuff is not yet ready to be used with the Java Module System as it causes illegal reflective access warnings.
Please make sure to switch the Java version in your IDE if you want to try it yourself.