Comparing py-java libraries

pyjava.py

# Example:
# PYJAVA_LIB=jpype venv/bin/python pyjava.py

import os
from datetime import datetime
from jpmml_evaluator import _package_classpath

lib = os.environ.get('PYJAVA_LIB')
assert lib is not None, 'Set env var PYJAVA_LIB to py4j/jnius/jpype'

##### Create a JVM
start_t = datetime.now()

if lib == 'py4j':
    from py4j.java_gateway import JavaGateway
    gateway = JavaGateway.launch_gateway(classpath=os.pathsep.join(_package_classpath()))
    jString = gateway.jvm.__getattr__('java.lang.String')
    jDouble = gateway.jvm.__getattr__('java.lang.Double')
    jFile = gateway.jvm.__getattr__('java.io.File')
    jLinkedHashMap = gateway.jvm.__getattr__('java.util.LinkedHashMap')
    jLoadingModelEvaluatorBuilder = gateway.jvm.__getattr__('org.jpmml.evaluator.LoadingModelEvaluatorBuilder')
    jModelEvaluationContext = gateway.jvm.__getattr__('org.jpmml.evaluator.ModelEvaluationContext')
    jEvaluatorUtil = gateway.jvm.__getattr__('org.jpmml.evaluator.EvaluatorUtil')

elif lib == 'jnius':
    import jnius_config
    jnius_config.set_classpath(*_package_classpath())
    import jnius
    jString = jnius.autoclass('java.lang.String')
    jDouble = jnius.autoclass('java.lang.Double')
    jFile = jnius.autoclass('java.io.File')
    jLinkedHashMap = jnius.autoclass('java.util.LinkedHashMap')
    jLoadingModelEvaluatorBuilder = jnius.autoclass('org.jpmml.evaluator.LoadingModelEvaluatorBuilder')
    jModelEvaluationContext = jnius.autoclass('org.jpmml.evaluator.ModelEvaluationContext')
    jEvaluatorUtil = jnius.autoclass('org.jpmml.evaluator.EvaluatorUtil')

elif lib == 'jpype':
    import jpype
    import jpype.imports
    jpype.startJVM(classpath=_package_classpath())
    from java.lang import String as jString
    from java.lang import Double as jDouble
    from java.io import File as jFile
    from java.util import LinkedHashMap as jLinkedHashMap
    from org.jpmml.evaluator import LoadingModelEvaluatorBuilder as jLoadingModelEvaluatorBuilder
    from org.jpmml.evaluator import ModelEvaluationContext as jModelEvaluationContext
    from org.jpmml.evaluator import EvaluatorUtil as jEvaluatorUtil

time = (datetime.now() - start_t).total_seconds()
print(f"createjvm: {time:.3f}s")

##### Load a Model
times = []
for _ in range(100):
    start_t = datetime.now()

    evaluatorBuilder = jLoadingModelEvaluatorBuilder()
    evaluatorBuilder.setLocatable(True)
    evaluatorBuilder.load(jFile(jString("jpmml_evaluator/tests/resources/DecisionTreeIris.pmml")))
    evaluator = evaluatorBuilder.build().verify()

    times.append((datetime.now() - start_t).total_seconds())
print(f"loadmodel: tot={sum(times):.6f} max={max(times):.6f}s avg={sum(times) / len(times):.6f}s")

##### Query field info
times = []
for _ in range(100):
    start_t = datetime.now()

    inputFields = evaluator.getInputFields()
    vals = [inputField.getName() for inputField in inputFields]
    # print("Input fields: ", vals)
    targetFields = evaluator.getTargetFields()
    vals = [targetField.getName() for targetField in targetFields]
    # print("Target field(s): ", vals)
    outputFields = evaluator.getOutputFields()
    vals = [outputField.getName() for outputField in outputFields]
    # print("Output fields: ", vals)

    times.append((datetime.now() - start_t).total_seconds())
print(f"fields   : tot={sum(times):.6f} max={max(times):.6f}s avg={sum(times) / len(times):.6f}s")

##### Score
times = []
for i in range(100):
    val = i * 0.001
    start_t = datetime.now()

    arguments1 = {
        "Sepal.Length" : 5.1 + val,
        "Sepal.Width" : 3.5 + val,
        "Petal.Length" : 1.4 + val,
        "Petal.Width" : 0.2 + val,
    }
    arguments2 = jLinkedHashMap()
    for k, v in arguments1.items():
        arguments2.put(jString(k), jDouble(v))
    arguments = jEvaluatorUtil.encodeKeys(arguments2)
    results1 = evaluator.evaluate(arguments)
    results2 = jEvaluatorUtil.decodeAll(results1)

    times.append((datetime.now() - start_t).total_seconds())
print(f"score    : tot={sum(times):.6f} max={max(times):.6f}s avg={sum(times) / len(times):.6f}s")

# jpype
# createjvm: 0.550s
# loadmodel: tot=1.466451 max=1.064521s avg=0.014665s
# fields   : tot=0.019881 max=0.009795s avg=0.000199s
# score    : tot=0.033356 max=0.023338s avg=0.000334s

# jnius
# createjvm: 0.249s
# loadmodel: tot=1.773011 max=1.385274s avg=0.017730s
# fields   : tot=0.039058 max=0.012234s avg=0.000391s
# score    : tot=0.067590 max=0.031904s avg=0.000676s

# py4j
# createjvm: 0.222s
# loadmodel: tot=0.616913 max=0.027464s avg=0.006169s
# fields   : tot=0.699152 max=0.026426s avg=0.006992s
# score    : tot=0.389583 max=0.017620s avg=0.003896s