public Model train(DataSet dataSet, SVMParameter parameter) throws OperatorException {
para = parameter;
setDataSourceInfo(
DataSourceInfoFactory.createConnectionInfo(
((DBTable) dataSet.getDBTable()).getDatabaseConnection().getProperties().getName()));
DatabaseConnection databaseConnection =
((DBTable) dataSet.getDBTable()).getDatabaseConnection();
Column label = dataSet.getColumns().getLabel();
String labelString = StringHandler.doubleQ(label.getName());
DataSet newDataSet = getTransformer().TransformCategoryToNumeric_new(dataSet);
String newTableName = ((DBTable) newDataSet.getDBTable()).getTableName();
Statement st = null;
ResultSet rs = null;
try {
st = databaseConnection.createStatement(false);
} catch (SQLException e) {
e.printStackTrace();
throw new OperatorException(e.getLocalizedMessage());
}
StringBuffer ind = getColumnArray(newDataSet);
StringBuffer where = getColumnWhere(newDataSet);
where.append(" and ").append(labelString).append(" is not null ");
SVMRegressionModel model = new SVMRegressionModel(dataSet, newDataSet);
if (!newDataSet.equals(dataSet)) {
model.setAllTransformMap_valueKey(getTransformer().getAllTransformMap_valueKey());
}
model.setKernelType(para.getKernelType());
model.setDegree(para.getDegree());
model.setGamma(para.getGamma());
String sql =
"select (model).inds, (model).cum_err, (model).epsilon, (model).rho, (model).b, (model).nsvs, (model).ind_dim, (model).weights, (model).individuals from (select alpine_miner_online_sv_reg('"
+ newTableName
+ "','"
+ ind
+ "','"
+ labelString
+ "','"
+ where
+ "',"
+ para.getKernelType()
+ ","
+ para.getDegree()
+ ","
+ para.getGamma()
+ ","
+ para.getEta()
+ ","
+ ((SVMRegressionParameter) para).getSlambda()
+ ","
+ para.getNu()
+ ") as model ";
if (getDataSourceInfo().getDBType().equals(DataSourceInfoOracle.dBType)) {
sql += " from dual ";
}
sql += ") a";
try {
itsLogger.debug("SVMRegression.train():sql=" + sql);
rs = st.executeQuery(sql.toString());
setModel(rs, model);
if (getTransformer().isTransform()) {
dropTable(st, newTableName);
}
rs.close();
st.close();
} catch (SQLException e) {
e.printStackTrace();
throw new OperatorException(e.getLocalizedMessage());
}
return model;
}
/* (non-Javadoc)
* @see com.alpine.datamining.api.impl.db.AbstractDBModelTrainer#train(com.alpine.datamining.api.AnalyticSource)
*/
@Override
protected Model train(AnalyticSource source) throws AnalysisException {
ResultSet rs = null;
Statement st = null;
EMModel trainModel = null;
try {
IDataSourceInfo dataSourceInfo =
DataSourceInfoFactory.createConnectionInfo(source.getDataSourceType());
dbtype = dataSourceInfo.getDBType();
EMConfig config = (EMConfig) source.getAnalyticConfig();
String anaColumns = config.getColumnNames();
String[] columnsArray = anaColumns.split(",");
List<String> transformColumns = new ArrayList<String>();
for (int i = 0; i < columnsArray.length; i++) {
transformColumns.add(columnsArray[i]);
}
DataSet dataSet = getDataSet((DataBaseAnalyticSource) source, config);
filerColumens(dataSet, transformColumns);
dataSet.computeAllColumnStatistics();
ColumnTypeTransformer transformer = new ColumnTypeTransformer();
DataSet newDataSet = transformer.TransformCategoryToNumeric_new(dataSet);
String tableName = ((DBTable) newDataSet.getDBTable()).getTableName();
Columns columns = newDataSet.getColumns();
List<String> newTransformColumns = new ArrayList<String>();
HashMap<String, String> transformMap = new HashMap<String, String>();
for (String key : transformer.getAllTransformMap_valueKey().keySet()) {
HashMap<String, String> values = (transformer.getAllTransformMap_valueKey()).get(key);
for (String lowKey : values.keySet()) {
transformMap.put(values.get(lowKey), lowKey);
}
}
Iterator<Column> attributeIter = columns.iterator();
while (attributeIter.hasNext()) {
Column column = attributeIter.next();
newTransformColumns.add(column.getName());
}
int maxIterationNumber = Integer.parseInt(config.getMaxIterationNumber());
int clusterNumber = Integer.parseInt(config.getClusterNumber());
double epsilon = Double.parseDouble(config.getEpsilon());
int initClusterSize = 10;
if (config.getInitClusterSize() != null) {
initClusterSize = Integer.parseInt(config.getInitClusterSize());
}
if (newDataSet.size() < initClusterSize * clusterNumber) {
initClusterSize = (int) (newDataSet.size() / clusterNumber + 1);
} // TODO get it from config and make sure it will not be too large
EMClusterImpl emImpl = EMClusterFactory.createEMAnalyzer(dbtype);
trainModel = EMClusterFactory.createEMModel(dbtype, newDataSet);
Connection connection = null;
connection = ((DataBaseAnalyticSource) source).getConnection();
st = connection.createStatement();
ArrayList<Double> tempResult =
emImpl.emTrain(
connection,
st,
tableName,
maxIterationNumber,
epsilon,
clusterNumber,
newTransformColumns,
initClusterSize,
trainModel);
trainModel = generateEMModel(trainModel, newTransformColumns, clusterNumber, tempResult);
if (!newDataSet.equals(this.dataSet)) {
trainModel.setAllTransformMap_valueKey(transformMap);
}
} catch (Exception e) {
logger.error(e);
if (e instanceof WrongUsedException) {
throw new AnalysisError(this, (WrongUsedException) e);
} else if (e instanceof AnalysisError) {
throw (AnalysisError) e;
} else {
throw new AnalysisException(e);
}
} finally {
try {
if (st != null) {
st.close();
}
if (rs != null) {
rs.close();
}
} catch (SQLException e) {
logger.debug(e.toString());
throw new AnalysisException(e.getLocalizedMessage());
}
}
return trainModel;
}