Awesome

JPMML-TensorFlow

Java library and command-line application for converting TensorFlow models to PMML.

Features

• Supported Estimator types:
  • learn.DNNClassifier
  • learn.DNNRegressor
  • learn.LinearClassifier
  • learn.LinearRegressor
• Supported Feature column types:
  • layers.one_hot_column
  • layers.real_valued_column
  • layers.sparse_column_with_keys
• Production quality:
  • Complete test coverage.
  • Fully compliant with the JPMML-Evaluator library.

Prerequisites

The TensorFlow side of operations

• Protocol Buffers 3.2.0 or newer
• TensorFlow 1.1.0 or newer

The Java side of operations

• Java 1.8 or newer

Installation

Enter the project root directory and build using Apache Maven; use the protoc.exe system property to specify the location of the Protocol Buffers compiler:

mvn -Dprotoc.exe=/usr/local/bin/protoc clean install

The build produces an executable uber-JAR file target/converter-executable-1.0-SNAPSHOT.jar.

Usage

A typical workflow can be summarized as follows:

1. Use TensorFlow to train an estimator.
2. Export the estimator in SavedModel data format to a directory in a local filesystem.
3. Use the JPMML-TensorFlow command-line converter application to turn the SavedModel directory to a PMML file.

The TensorFlow side of operations

Please see the test script file main.py for sample workflows.

The Java side of operations

Converting the estimator SavedModel directory estimator/ to a PMML file estimator.pmml:

java -jar target/converter-executable-1.0-SNAPSHOT.jar --tf-savedmodel-input estimator/ --pmml-output estimator.pmml

Getting help:

java -jar target/converter-executable-1.0-SNAPSHOT.jar --help

License

JPMML-TensorFlow is licensed under the GNU Affero General Public License (AGPL) version 3.0. Other licenses are available on request.

Additional information

Please contact info@openscoring.io