weka.classifiers.sklearn.ScikitLearnClassifier Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of wekaPython Show documentation
Show all versions of wekaPython Show documentation
Integration with CPython for Weka. Python version 2.7.x or higher is required. Also requires the following packages to be installed in python: numpy, pandas, matplotlib and scikit-learn. This package provides a wrapper classifier and clusterer that, between them, cover 60+ scikit-learn algorithms. It also provides a general scripting step for the Knowlege Flow along with scripting plugin environments for the Explorer and Knowledge Flow.
The newest version!
/*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see outAndErr = session.executeScript(script, getDebug());
String versionNumber =
session.getVariableValueFromPythonAsPlainString("skv", getDebug());
if (versionNumber != null && versionNumber.length() > 0) {
// strip minor version
versionNumber = versionNumber.substring(0, versionNumber.lastIndexOf('.'));
m_scikitVersion = Double.parseDouble(versionNumber);
}
// check for xgboost
m_xgboostInstalled = true;
script = "import xgboost\n";
outAndErr = session.executeScript(script, getDebug());
if (outAndErr.get(1).length() > 0) {
m_xgboostInstalled = false;
}
} catch (WekaException e) {
m_xgboostInstalled = false;
e.printStackTrace();
} finally {
// PythonSession.releaseSession(this);
try {
releaseSession();
session = null;
} catch (WekaException e) {
e.printStackTrace();
}
}
}
result.enable(Capabilities.Capability.NUMERIC_ATTRIBUTES);
result.enable(Capabilities.Capability.NOMINAL_ATTRIBUTES);
result.enable(Capabilities.Capability.DATE_ATTRIBUTES);
result.enable(Capabilities.Capability.MISSING_VALUES);
result.enable(Capabilities.Capability.MISSING_CLASS_VALUES);
if (m_learner.isClassifier()) {
result.enable(Capabilities.Capability.BINARY_CLASS);
result.enable(Capabilities.Capability.NOMINAL_CLASS);
}
if (m_learner.isRegressor()) {
result.enable(Capabilities.Capability.NUMERIC_CLASS);
}
if (session != null) {
try {
releaseSession();
} catch (WekaException e) {
e.printStackTrace();
}
}
} else {
try {
String pyCheckResults = getPythonEnvCheckResults();
throw new RuntimeException(String.format("The python environment is either not available or is not configured correctly:\n\n%s", pyCheckResults));
} catch (WekaException e) {
throw new RuntimeException("The python environment is either not available or is not configured correctly", e);
}
}
} else {
result.enable(Capabilities.Capability.NUMERIC_ATTRIBUTES);
result.enable(Capabilities.Capability.NOMINAL_ATTRIBUTES);
result.enable(Capabilities.Capability.DATE_ATTRIBUTES);
result.enable(Capabilities.Capability.MISSING_VALUES);
result.enable(Capabilities.Capability.MISSING_CLASS_VALUES);
if (m_learner.isClassifier()) {
result.enable(Capabilities.Capability.BINARY_CLASS);
result.enable(Capabilities.Capability.NOMINAL_CLASS);
}
if (m_learner.isRegressor()) {
result.enable(Capabilities.Capability.NUMERIC_CLASS);
}
}
return result;
}
/**
* Get the capabilities of this learner
*
* @return the capabilities of this scheme
*/
@Override
public Capabilities getCapabilities() {
// note that we don't do the server checks from here because the GenericObjectEditor
// in the GUI calls this method for every character typed in a text field!
return getCapabilities(false);
}
/**
* Get whether to use the supervised version of nominal to binary
*
* @return true if supervised nominal to binary is to be used
*/
@OptionMetadata(displayName = "Use supervised nominal to binary conversion",
description = "Use supervised nominal to binary conversion of nominal attributes.",
commandLineParamName = "S", commandLineParamSynopsis = "-S",
commandLineParamIsFlag = true, displayOrder = 3)
public boolean getUseSupervisedNominalToBinary() {
return m_useSupervisedNominalToBinary;
}
/**
* Set whether to use the supervised version of nominal to binary
*
* @param useSupervisedNominalToBinary true if supervised nominal to binary is
* to be used
*/
public void
setUseSupervisedNominalToBinary(boolean useSupervisedNominalToBinary) {
m_useSupervisedNominalToBinary = useSupervisedNominalToBinary;
}
/**
* Get the scikit-learn scheme to use
*
* @return the scikit-learn scheme to use
*/
@OptionMetadata(displayName = "Scikit-learn learner",
description = "Scikit-learn learner to use.\nAvailable learners:\nDecisionTreeClassifier"
+ ", DecisionTreeRegressor, GaussianNB, MultinomialNB,"
+ "BernoulliNB, LDA, QDA, "
+ "LogisticRegression, LogisticRegressionCV,\n"
+ "LinearRegression, ARDRegression, "
+ "BayesianRidge, ElasticNet, Lars,\nLarsCV, Lasso, LassoCV, LassoLars, "
+ "LassoLarsCV, LassoLarsIC, MLPClassifier, MLPRegressor, OrthogonalMatchingPursuit,\n"
+ "OrthogonalMatchingPursuitCV, PassiveAggressiveClassifier, "
+ "PassiveAggressiveRegressor, Perceptron, RANSACRegressor,\nRidge, "
+ "RidgeClassifier, RidgeClassifierCV, RidgeCV, SGDClassifier,\nSGDRegressor,"
+ "TheilSenRegressor, GaussianProcess, KernelRidge, KNeighborsClassifier, "
+ "\nRadiusNeighborsClassifier, KNeighborsRegressor, RadiusNeighborsRegressor, SVC,"
+ "\nLinearSVC, NuSVC, SVR, NuSVR, AdaBoostClassifier, AdaBoostRegressor,"
+ "BaggingClassifier, BaggingRegressor,\nExtraTreeClassifier, ExtraTreeRegressor,"
+ "GradientBoostingClassifier, GradientBoostingRegressor,\n"
+ "RandomForestClassifier, RandomForestRegressor, XGBClassifier, XGBRegressor."
+ "\n(default = DecisionTreeClassifier)",
commandLineParamName = "learner",
commandLineParamSynopsis = "-learner ", displayOrder = 1)
public SelectedTag getLearner() {
return new SelectedTag(m_learner.ordinal(), TAGS_LEARNER);
}
/**
* Set the scikit-learn scheme to use
*
* @param learner the scikit-learn scheme to use
*/
public void setLearner(SelectedTag learner) {
int learnerID = learner.getSelectedTag().getID();
for (Learner l : Learner.values()) {
if (l.ordinal() == learnerID) {
m_learner = l;
break;
}
}
}
/**
* Get the parameters to pass to the scikit-learn scheme
*
* @return the parameters to use
*/
@OptionMetadata(displayName = "Learner parameters",
description = "learner parameters to use", displayOrder = 2,
commandLineParamName = "parameters",
commandLineParamSynopsis = "-parameters ")
public String getLearnerOpts() {
return m_learnerOpts;
}
/**
* Set the parameters to pass to the scikit-learn scheme
*
* @param opts the parameters to use
*/
public void setLearnerOpts(String opts) {
m_learnerOpts = opts;
}
@Override
public void setBatchSize(String size) {
m_batchPredictSize = size;
}
@OptionMetadata(displayName = "Batch size", description = "The preferred "
+ "number of instances to transfer into python for prediction\n(if operating"
+ "in batch prediction mode). More or fewer instances than this will be "
+ "accepted.", commandLineParamName = "batch",
commandLineParamSynopsis = "-batch ", displayOrder = 4)
@Override
public String getBatchSize() {
return m_batchPredictSize;
}
/**
* Returns true, as we send entire test sets over to python for prediction
*
* @return true
*/
@Override
public boolean implementsMoreEfficientBatchPrediction() {
return true;
}
/**
* Set whether to try and continue after seeing output on the sys error
* stream. Some schemes write warnings (rather than errors) to sys error.
*
* @param c true if we should try to continue after seeing output on the sys
* error stream
*/
public void setContinueOnSysErr(boolean c) {
m_continueOnSysErr = c;
}
/**
* Get whether to try and continue after seeing output on the sys error
* stream. Some schemes write warnings (rather than errors) to sys error.
*
* @return true if we should try to continue after seeing output on the sys
* error stream
*/
@OptionMetadata(
displayName = "Try to continue after sys err output from script",
description = "Try to continue after sys err output from script.\nSome schemes"
+ " report warnings to the system error stream.",
displayOrder = 5, commandLineParamName = "continue-on-err",
commandLineParamSynopsis = "-continue-on-err",
commandLineParamIsFlag = true)
public boolean getContinueOnSysErr() {
return m_continueOnSysErr;
}
/**
* If true then don't retrieve the model from python. This can speed up
* cross-validation but will prevent the trained classifier from being used
* after deserialization. May also be necessary if a model in python is very
* large and exceeds the current transfer size (Integer.MAX_VALUE bytes, about
* 2.1Gb).
*
* @param dontFetchModelFromPython true to not fetch the model from python.
*/
public void setDontFetchModelFromPython(boolean dontFetchModelFromPython) {
m_dontFetchModelFromPython = dontFetchModelFromPython;
}
/**
* If true then don't retrieve the model from python. This can speed up
* cross-validation but will prevent the trained classifier from being used
* after deserialization. May also be necessary if a model in python is very
* large and exceeds the current transfer size (Integer.MAX_VALUE bytes, about
* 2.1Gb).
*
* @return true if not fetching the model from python.
*/
@OptionMetadata(displayName = "Don't retrieve model from python",
description = "Don't retrieve the model from python - speeds up "
+ "cross-validation,\nbut prevents this classifier from being "
+ "used after deserialization.\nSome models in python (e.g. large "
+ "random forests) may exceed the maximum size for transfer\n("
+ "currently Integer.MAX_VALUE bytes)",
displayOrder = 6, commandLineParamName = "dont-fetch-model",
commandLineParamSynopsis = "-dont-fetch-model",
commandLineParamIsFlag = true)
public boolean getDontFetchModelFromPython() {
return m_dontFetchModelFromPython;
}
/**
* Set the python command to use. Empty string or "default" indicate that the
* python present in the PATH should be used.
*
* @param pyCommand the path to the python executable (or empty
* string/"default" to use python in the PATH)
*/
public void setPythonCommand(String pyCommand) {
m_pyCommand = pyCommand;
}
/**
* Get the python command to use. Empty string or "default" indicate that the
* python present in the PATH should be used.
*
* @return the path to the python executable (or empty string/"default" to use
* python in the PATH)
*/
@OptionMetadata(displayName = "Python command ",
description = "Path to python executable ('default' to use python in the PATH)",
commandLineParamName = "py-command",
commandLineParamSynopsis = "-py-command ",
displayOrder = 7)
public String getPythonCommand() {
return m_pyCommand;
}
/**
* Set optional entries to prepend to the PATH so that python can execute
* correctly. Only applies when not using a default python.
*
* @param pythonPath additional entries to prepend to the PATH
*/
public void setPythonPath(String pythonPath) {
m_pyPath = pythonPath;
}
/**
* Get optional entries to prepend to the PATH so that python can execute
* correctly. Only applies when not using a default python.
*
* @return additional entries to prepend to the PATH
*/
@OptionMetadata(displayName = "Python path",
description = "Optional elements to prepend to the PATH so that python can "
+ "execute correctly ('default' to use PATH as-is)",
commandLineParamName = "py-path",
commandLineParamSynopsis = "-py-path ", displayOrder = 8)
public String getPythonPath() {
return m_pyPath;
}
/**
* Set an optional server name by which to identify the python server to use.
* Can be used share a given server amongst selected instances or reserve a
* server specifically for this classifier. Python command + server name
* uniquely identifies the server to use.
*
* @param serverID the name of the server to use (or none for no specific
* server name).
*/
public void setServerID(String serverID) {
m_serverID = serverID;
}
/**
* Get an optional server name by which to identify the python server to use.
* Can be used share a given server amongst selected instances or reserve a
* server specifically for this classifier. Python command + server name
* uniquely identifies the server to use.
*
* @return the name of the server to use (or none) for no specific server
* name.
*/
@OptionMetadata(displayName = "Server name/ID",
description = "Optional name to identify this server, can be used to share "
+ "a given server instance - default = 'none' (i.e. no server name)",
commandLineParamName = "server-name",
commandLineParamSynopsis = "-server-name ",
displayOrder = 9)
public String getServerID() {
return m_serverID;
}
/**
* Gets a python session object to use for interacting with python
*
* @return a PythonSession object
* @throws WekaException if a problem occurs
*/
protected PythonSession getSession() throws WekaException {
PythonSession session = null;
String pyCommand = m_pyCommand != null && m_pyCommand.length() > 0
&& !m_pyCommand.equalsIgnoreCase("default") ? m_pyCommand : null;
String pyPath = m_pyPath != null && m_pyPath.length() > 0
&& !m_pyPath.equalsIgnoreCase("default") ? m_pyPath : null;
String serverID = m_serverID != null && m_serverID.length() > 0
&& !m_serverID.equalsIgnoreCase("none") ? m_serverID : null;
if (pyCommand != null) {
if (!PythonSession.pythonAvailable(pyCommand, serverID)) {
// try to create this environment/server
// System.err.println("Starting server: " + pyCommand + " " + serverID);
if (!PythonSession.initSession(pyCommand, serverID, pyPath, getDebug())) {
String envEvalResults =
PythonSession.getPythonEnvCheckResults(pyCommand, serverID);
throw new WekaException("Was unable to start python environment ("
+ pyCommand + ")\n\n" + envEvalResults);
}
}
session = PythonSession.acquireSession(pyCommand, serverID, this);
} else {
if (!PythonSession.pythonAvailable()) {
// try initializing
if (!PythonSession.initSession("python", getDebug())) {
String envEvalResults = PythonSession.getPythonEnvCheckResults();
throw new WekaException(
"Was unable to start python environment:\n\n" + envEvalResults);
}
}
session = PythonSession.acquireSession(this);
}
return session;
}
/**
* Release the python session
*
* @throws WekaException if a problem occurs
*/
protected void releaseSession() throws WekaException {
PythonSession session = null;
String pyCommand = m_pyCommand != null && m_pyCommand.length() > 0
&& !m_pyCommand.equalsIgnoreCase("default") ? m_pyCommand : null;
String serverID = m_serverID != null && m_serverID.length() > 0
&& !m_serverID.equalsIgnoreCase("none") ? m_serverID : null;
if (pyCommand != null) {
if (PythonSession.pythonAvailable(pyCommand, serverID)) {
PythonSession.releaseSession(pyCommand, serverID, this);
}
} else {
if (PythonSession.pythonAvailable()) {
PythonSession.releaseSession(this);
}
}
}
/**
* Get the results of executing the environment check in python
*
* @return the results of executing the environment check
* @throws WekaException if a problem occurs
*/
protected String getPythonEnvCheckResults() throws WekaException {
String result = "";
String pyCommand = m_pyCommand != null && m_pyCommand.length() > 0
&& !m_pyCommand.equalsIgnoreCase("default") ? m_pyCommand : null;
String serverID = m_serverID != null && m_serverID.length() > 0
&& !m_serverID.equalsIgnoreCase("none") ? m_serverID : null;
if (pyCommand != null) {
result = PythonSession.getPythonEnvCheckResults(pyCommand, serverID);
} else {
result = PythonSession.getPythonEnvCheckResults();
}
return result;
}
/**
* Check to see if python is available for the user-specified environment.
*
* @return
*/
protected boolean pythonAvailable() {
String pyCommand = m_pyCommand != null && m_pyCommand.length() > 0
&& !m_pyCommand.equalsIgnoreCase("default") ? m_pyCommand : null;
String serverID = m_serverID != null && m_serverID.length() > 0
&& !m_serverID.equalsIgnoreCase("none") ? m_serverID : null;
if (pyCommand != null) {
return PythonSession.pythonAvailable(pyCommand, serverID);
}
return PythonSession.pythonAvailable();
}
/**
* Build the classifier
*
* @param data set of instances serving as training data
* @throws Exception if a problem occurs
*/
@Override
public void buildClassifier(Instances data) throws Exception {
m_pickledModel = null;
getCapabilities(true).testWithFail(data);
m_zeroR = null;
PythonSession session = getSession();
if (m_learner == Learner.XGBClassifier && !m_xgboostInstalled) {
throw new Exception(
"xgboost does not seem to be available in your python "
+ "installation");
}
if (m_modelHash == null) {
m_modelHash = "" + hashCode();
}
data = new Instances(data);
data.deleteWithMissingClass();
m_zeroR = new ZeroR();
m_zeroR.buildClassifier(data);
m_classPriors = data.numInstances() > 0
? m_zeroR.distributionForInstance(data.instance(0))
: new double[data.classAttribute().numValues()];
if (data.numInstances() == 0 || data.numAttributes() == 1) {
if (data.numInstances() == 0) {
System.err
.println("No instances with non-missing class - using ZeroR model");
} else {
System.err.println("Only the class attribute is present in "
+ "the data - using ZeroR model");
}
return;
} else {
m_zeroR = null;
}
if (data.classAttribute().isNominal()) {
// check for empty classes
AttributeStats stats = data.attributeStats(data.classIndex());
m_nominalEmptyClassIndexes =
new boolean[data.classAttribute().numValues()];
for (int i = 0; i < stats.nominalWeights.length; i++) {
if (stats.nominalWeights[i] == 0) {
m_nominalEmptyClassIndexes[i] = true;
}
}
}
m_replaceMissing.setInputFormat(data);
data = Filter.useFilter(data, m_replaceMissing);
if (getUseSupervisedNominalToBinary()) {
m_nominalToBinary =
new weka.filters.supervised.attribute.NominalToBinary();
} else {
m_nominalToBinary =
new weka.filters.unsupervised.attribute.NominalToBinary();
}
m_nominalToBinary.setInputFormat(data);
data = Filter.useFilter(data, m_nominalToBinary);
try {
String learnerModule = m_learner.getModule();
String learnerMethod = m_learner.toString();
if (learnerMethod.equalsIgnoreCase("MLPClassifier")
|| learnerMethod.equalsIgnoreCase("MLPRegressor")) {
if (m_scikitVersion < 0.18) {
throw new Exception(
learnerMethod + " is not available in scikit-learn " + "version "
+ m_scikitVersion + ". Version 0.18 or higher is " + "required.");
}
}
// PythonSession session = PythonSession.acquireSession(this);
// transfer the data over to python
session.instancesToPythonAsScikitLearn(data, TRAINING_DATA_ID,
getDebug());
StringBuilder learnScript = new StringBuilder();
if (m_learner == Learner.XGBClassifier
|| m_learner == Learner.XGBRegressor) {
learnScript.append("import xgboost as xgb\n")
.append("import numpy as np").append("\n");
} else {
learnScript.append("from sklearn import *").append("\n")
.append("import numpy as np").append("\n");
}
if (m_scikitVersion > 0.18) {
if (learnerMethod.equalsIgnoreCase("LDA")) {
learnerMethod = "LinearDiscriminantAnalysis";
learnerModule = "discriminant_analysis";
} else if (learnerMethod.equalsIgnoreCase("QDA")) {
learnerMethod = "QuadraticDiscriminantAnalysis";
learnerModule = "discriminant_analysis";
}
}
if (m_learner == Learner.XGBClassifier
|| m_learner == Learner.XGBRegressor) {
learnScript.append(MODEL_ID + m_modelHash + " = xgb." + learnerMethod
+ "(" + getLearnerOpts() + ")").append("\n");
} else {
learnScript
.append(MODEL_ID + m_modelHash + " = " + learnerModule + "."
+ learnerMethod + "("
+ (getLearnerOpts().length() > 0 ? getLearnerOpts() : "") + ")")
.append("\n");
}
learnScript.append(MODEL_ID + m_modelHash + ".fit(X,np.ravel(Y))")
.append("\n");
List outAndErr =
session.executeScript(learnScript.toString(), getDebug());
if (outAndErr.size() == 2 && outAndErr.get(1).length() > 0) {
if (m_continueOnSysErr) {
System.err.println(outAndErr.get(1));
} else {
throw new Exception(outAndErr.get(1));
}
}
m_learnerToString = session.getVariableValueFromPythonAsPlainString(
MODEL_ID + m_modelHash, getDebug());
if (!getDontFetchModelFromPython()) {
// retrieve the model from python
m_pickledModel = session.getVariableValueFromPythonAsPickledObject(
MODEL_ID + m_modelHash, getDebug());
}
if (m_learner.removeModelFromPythonPostTrainPredict()) {
String cleanUp = "del " + MODEL_ID + m_modelHash + "\n";
outAndErr = session.executeScript(cleanUp, getDebug());
if (outAndErr.size() == 2 && outAndErr.get(1).length() > 0) {
if (m_continueOnSysErr) {
System.err.println(outAndErr.get(1));
} else {
throw new Exception(outAndErr.get(1));
}
}
}
// release session
} finally {
releaseSession();
}
}
private double[][] batchScoreWithZeroR(Instances insts) throws Exception {
double[][] result = new double[insts.numInstances()][];
for (int i = 0; i < insts.numInstances(); i++) {
Instance current = insts.instance(i);
result[i] = m_zeroR.distributionForInstance(current);
}
return result;
}
/**
* Return the predicted probabilities for the supplied instance
*
* @param instance the instance to be classified
* @return the predicted probabilities
* @throws Exception if a problem occurs
*/
@Override
public double[] distributionForInstance(Instance instance) throws Exception {
Instances temp = new Instances(instance.dataset(), 0);
temp.add(instance);
return distributionsForInstances(temp)[0];
}
/**
* Return the predicted probabilities for the supplied instances
*
* @param insts the instances to get predictions for
* @return the predicted probabilities for the supplied instances
* @throws Exception if a problem occurs
*/
@Override
@SuppressWarnings("unchecked")
public double[][] distributionsForInstances(Instances insts)
throws Exception {
if (m_zeroR != null) {
return batchScoreWithZeroR(insts);
}
insts = Filter.useFilter(insts, m_replaceMissing);
insts = Filter.useFilter(insts, m_nominalToBinary);
Attribute classAtt = insts.classAttribute();
// remove the class attribute
Remove r = new Remove();
r.setAttributeIndices("" + (insts.classIndex() + 1));
r.setInputFormat(insts);
insts = Filter.useFilter(insts, r);
insts.setClassIndex(-1);
double[][] results = null;
PythonSession session = null;
try {
session = getSession();
session.instancesToPythonAsScikitLearn(insts, TEST_DATA_ID, getDebug());
StringBuilder predictScript = new StringBuilder();
// check if model exists in python. If not, then transfer it over
if (!session.checkIfPythonVariableIsSet(MODEL_ID + m_modelHash,
getDebug())) {
if (m_pickledModel == null || m_pickledModel.length() == 0) {
throw new Exception("There is no model to transfer into Python!");
}
session.setPythonPickledVariableValue(MODEL_ID + m_modelHash,
m_pickledModel, getDebug());
}
String learnerModule = m_learner.getModule();
String learnerMethod = m_learner.toString();
if (m_scikitVersion > 0.18) {
if (learnerMethod.equalsIgnoreCase("LDA")) {
learnerMethod = "LinearDiscriminantAnalysis";
learnerModule = "discriminant_analysis";
} else if (learnerMethod.equalsIgnoreCase("QDA")) {
learnerMethod = "QuadraticDiscriminantAnalysis";
learnerModule = "discriminant_analysis";
}
}
if (m_learner == Learner.XGBClassifier
|| m_learner == Learner.XGBRegressor) {
predictScript.append("import xgboost as xgb\n");
} else {
predictScript
.append("from sklearn." + learnerModule + " import " + learnerMethod)
.append("\n");
}
predictScript.append("preds = " + MODEL_ID + m_modelHash + ".predict"
+ (m_learner.producesProbabilities(m_learnerOpts) ? "_proba" : "")
+ "(X)").append("\npreds = preds.tolist()\n");
List outAndErr =
session.executeScript(predictScript.toString(), getDebug());
if (outAndErr.size() == 2 && outAndErr.get(1).length() > 0) {
if (m_continueOnSysErr) {
System.err.println(outAndErr.get(1));
} else {
throw new Exception(outAndErr.get(1));
}
}
List © 2015 - 2024 Weber Informatics LLC | Privacy Policy