Time Series Parameter learning in Python

This example makes use of the Python code in Data Frame Utils.

# __author__ = 'Bayes Server'
# __version__= '0.2'

import pandas as pd
import jpype    # pip install jpype1    (version 1.2.1 or later)
import jpype.imports
from jpype.types import *
from jpype import java, JImplements, JOverride

classpath = "lib/bayesserver-10.8.jar"  # TODO download the Bayes Server Java API, and adjust the path

# Launch the JVM
jpype.startJVM(classpath=[classpath])

import data_frame_utils as dfu

# import the Java modules
from com.bayesserver import *
from com.bayesserver.inference import *
from com.bayesserver.data import *
from com.bayesserver.learning.parameters import *

# Uncomment the following line and change the license key, if you are using a licensed version
# License.validate("xxx")


@JImplements(ParameterLearningProgress)
class ConsoleParameterLearningProgress(object):

    @JOverride
    def update(self, info):
        """
        See ParameterLearningProgress in the Bayes Server .NET or Java API docs for details.
        """
        print('Iterations: ', info.getIterationCount(), ', log-likelihood: ', info.getLogLikelihood())

    @JOverride
    def getDistributionMonitoring(self):
        """
        See ParameterLearningProgress in the Bayes Server .NET or Java API docs for details.
        """
        return DistributionMonitoring.NONE


def create_network_structure():

    network = Network()

    x1 = Variable('X1', VariableValueType.CONTINUOUS)
    x2 = Variable('X2', VariableValueType.CONTINUOUS)

    # add a temporal (time series) node, with two continuous variables
    node_x = Node("X", [x1, x2])
    node_x.setTemporalType(TemporalType.TEMPORAL)

    network.getNodes().add(node_x)

    # add temporal links
    for order in range(1, 4):
        network.getLinks().add(Link(node_x, node_x, order))

    # at this point the Dynamic Bayesian network structure is fully specified

    return network


def learn_parameters():

    # we manually construct the network here, but it could be loaded from a file

    network = create_network_structure()

    x1 = network.getVariables().get("X1")
    x2 = network.getVariables().get("X2")

    # now learn the parameters from the data in Walkthrough 3 - Time Series network

    # This example uses Sql Server as the data source and assumes the data has been copied to
    # a table called TimeSeriesWalkthrough
    # We will use the RelevanceTree algorithm here, as it is optimized for parameter learning

    learning = ParameterLearning(network, RelevanceTreeInferenceFactory())
    learning_options = ParameterLearningOptions()
    learning_options.setTimeSeriesMode(TimeSeriesMode.ROLLING)
    progress = ConsoleParameterLearningProgress()
    learning_options.setProgress(progress)

    df = pd.read_csv('data/time_series_tutorial.csv')  #  Note: Saved as csv from the Tutorial data installed with Bayes Server
    dt = dfu.to_data_table(df)
    temporal_data_reader_command = DataTableDataReaderCommand(dt)

    temporal_reader_options = TemporalReaderOptions('Case', 'Time', TimeValueType.INDEX)

    # here we map variables to database columns
    # in this case the variables and database columns have the same name
    temporal_variable_references = [
                        VariableReference(x1, ColumnValueType.VALUE, x1.getName()),
                        VariableReference(x2, ColumnValueType.VALUE, x2.getName())
                ]

    # note that although this example only has temporal data
    # we could have included additional non temporal variables and data

    evidence_reader_command = DefaultEvidenceReaderCommand(
                temporal_data_reader_command,
                java.util.Arrays.asList(temporal_variable_references),
                temporal_reader_options)

    result = learning.learn(evidence_reader_command, learning_options)

    print("Log likelihood = ", result.getLogLikelihood())


learn_parameters()