Generate and Evaluation PMML in Python. Currently, this pacakge supports the Gaussian Process Regression and DeepNeuralNetwork model types.

All code is written in Python 3. First install the required packages

conda env create -f environment.yml -n pmml3
conda activate pmml3

Then run the tests to make sure everything is working

python tests.py

If all the tests pass, then you are good to start using the package.

Evaluating a PMML file with the commandline is straightforward. Models inputs are defined in json or image files.

# Gaussian process regression evaluation
python main.py predict \
	--model=examples/gpr/energy-prediction-1.pmml \
	--input=test/assets/energy-inputs.json

# Image classification with a deep neural network
python main.py predict \
    --model=examples/deepnetwork/VGG16/model.pmml \
    --input=test/assets/cat.jpg

Any PMML file can be validated against the DeepNetwork schema:

# Validate examples/deepnetwork/VGG16/model.pmml
python main.py validate --filename=examples/deepnetwork/VGG16/model.pmml

# Validate all of the examples
python main.py validate

A scoring engine server can also be started from a model file. Inputs to the scoring engine can also be sent as image files or json.

# Gaussian process regression evaluation scoring engine
python main.py runserver \
   --model=examples/gpr/energy-prediction-1.pmml \
   --port=5000

# Neural network scoring engine
python main.py runserver \
	--model=examples/deepnetwork/VGG16/model.pmml \
	--port=5000 \

Most of the examples models have been generated from other open-source projects. To regenerate the example models:

# Build new PMML files
python main.py build_keras_examples
python main.py build_pytorch_examples

Queries can be sent to the scoring engine server using standard JSON:

#!Python3
r = HTTP.get({'xnew': [1,4,5,3,5,7,8,4,3,6,7,1]})
r.response # -> {mu:1.45324344,sd:3.2214342}

The build hash in not included in the dependencies, making it easier to install the environment on different platforms. To export the environment:

conda env export --no-builds | grep -v "prefix" > environment.yml

• Max Ferguson: @maxkferg
• Stanford Engineering Informatics Group: eil.stanford.edu