PMML evaluator library for the [PostgreSQL database] (http://www.postgresql.org/).
- Full support for PMML specification versions 3.0 through 4.2. The evaluation is handled by the [JPMML-Evaluator] (https://github.com/jpmml/jpmml-evaluator) library.
- PostgreSQL version 8.X or 9.X database with a [PL/Java add-on module] (https://github.com/tada/pljava).
A working JPMML-PostgreSQL setup consists of a library JAR file and a number of model JAR files. The library JAR file is centered around the utility class org.jpmml.postgresql.PMMLUtil, which provides PL/Java compliant utility methods for handling most common PMML evaluation scenarios. A model JAR file contains a model launcher class and a corresponding PMML resource. Model JAR files can be added, replaced and/or removed on the go using PL/Java SQL commands. All changes take effect immediately. There is no need to modify PostgreSQL database configuration or restart the service.
The main responsibility of a model launcher class is to formalize the "public interface" of a PMML resource. Every PL/Java function must be backed by a public static method that takes a PostgreSQL tuple as an argument, and returns either a PostgreSQL scalar or a PostgreSQL tuple as a result.
The example model JAR file contains a DecisionTree model for the "iris" dataset. This model is exposed in two ways. First, the model launcher class org.jpmml.postgresql.DecisionTreeIris defines two functions that return the PMML target field ("Species") together with four output fields ("Predicted_Species", "Probability_setosa", "Probability_versicolor", "Probability_virginica") as a PostgreSQL tuple. Second, the model launcher class org.jpmml.postgresql.DecisionTreeIris_Species defines two functions that return the PMML target field ("Species") as a PostgreSQL character type. The installation and removal of functions is completely automated using an SQL deployment descriptor mechanism.
Enter the project root directory and build using [Apache Maven] (http://maven.apache.org/):
mvn clean install
The build produces two JAR files:
pmml-postgresql/target/pmml-postgresql-runtime-1.0-SNAPSHOT.jar- Library uber-JAR file. It contains the classes of the library JAR filepmml-postgresql/target/pmml-postgresql-1.0-SNAPSHOT.jar, plus all the classes of its transitive dependencies.pmml-postgresql-example/target/pmml-postgresql-example-1.0-SNAPSHOT.jar- Example model JAR file.
Install the library uber-JAR file:
SELECT sqlj.install_jar('file:///tmp/pmml-postgresql-runtime-1.0-SNAPSHOT.jar', 'jpmml', false);The PL/Java function sqlj.install_jar takes three arguments:
- The URL of the JAR file.
- A symbolic name after which this JAR file is later known for.
- A flag indicating if this JAR file contains an executable SQL deployment descriptor ("BEGIN INSTALL ...").
Add the library uber-JAR file to the classpath of the target schema:
SELECT sqlj.set_classpath('public', 'jpmml');Remove the library uber-JAR file:
SELECT sqlj.remove_jar('jpmml', false);The PL/Java function sqlj.remove_jar takes two arguments:
- The symbolic name of the JAR file.
- A flag indicating if this JAR file contains an executable SQL deployment descriptor ("BEGIN REMOVE ...").
PL/Java examines all classpaths and propagates the removal if appropriate.
Install the example model JAR file:
SELECT sqlj.install_jar('file:///tmp/pmml-postgresql-example-1.0-SNAPSHOT.jar', 'DecisionTreeIris', true);Behind the scenes, the SQL deployment descriptor orders the creation of two composite types and two functions as follows:
CREATE TYPE iris_request AS (
"Sepal_Length" double precision,
"Sepal_Width" double precision,
"Petal_Length" double precision,
"Petal_Width" double precision
);
CREATE TYPE iris_response AS (
"Species" varchar,
"Predicted_Species" varchar,
"Probability_setosa" double precision,
"Probability_versicolor" double precision,
"Probability_virginica" double precision
);
CREATE FUNCTION DecisionTreeIris(iris_request) RETURNS iris_response
AS 'org.jpmml.postgresql.DecisionTreeIris.evaluate'
LANGUAGE java;
CREATE FUNCTION DecisionTreeIris("Sepal_Length" double precision, "Sepal_Width" double precision, "Petal_Length" double precision, "Petal_Width" double precision) RETURNS iris_response
AS 'org.jpmml.postgresql.DecisionTreeIris.evaluate'
LANGUAGE java;
CREATE FUNCTION DecisionTreeIris_Species(iris_request) RETURNS varchar
AS 'org.jpmml.postgresql.DecisionTreeIris_Species.evaluate'
LANGUAGE java;
CREATE FUNCTION DecisionTreeIris_Species("Sepal_Length" double precision, "Sepal_Width" double precision, "Petal_Length" double precision, "Petal_Width" double precision) RETURNS varchar
AS 'org.jpmml.postgresql.DecisionTreeIris_Species.evaluate'
LANGUAGE java;Add the example model JAR file to the classpath of the target schema. The classpath is constructed by concatenating the symbolic name of the library uber-JAR file with the symbolic names of model JAR files (using comma : as a path separator character):
SELECT sqlj.set_classpath('public', 'jpmml:DecisionTreeIris');Predicting the iris species together with the calculated probabilities for each target category:
SELECT (DecisionTreeIris(7, 3.2, 4.7, 1.4)).*;Result:
Species | Predicted_Species | Probability_setosa | Probability_versicolor | Probability_virginica
------------+-------------------+--------------------+------------------------+-----------------------
versicolor | versicolor | 0 | 0.907407 | 0.0925926
Predicting the iris species:
SELECT DecisionTreeIris_Species(7, 3.2, 4.7, 1.4);Result:
decisiontreeiris_species
-------------------------
versicolor
Remove the example model JAR file:
SELECT sqlj.remove_jar('DecisionTreeIris', true);Behind the scenes, the SQL deployment descriptor orders the deletion of two composite types and two functions as follows:
DROP FUNCTION DecisionTreeIris_Species(double precision, double precision, double precision, double precision);
DROP FUNCTION DecisionTreeIris_Species(iris_request);
DROP FUNCTION DecisionTreeIris(double precision, double precision, double precision, double precision);
DROP FUNCTION DecisionTreeIris(iris_request);
DROP TYPE iris_response;
DROP TYPE iris_request;JPMML-PostgreSQL is dual-licensed under the [GNU Affero General Public License (AGPL) version 3.0] (http://www.gnu.org/licenses/agpl-3.0.html) and a commercial license.
Please contact [info@openscoring.io] (mailto:info@openscoring.io)