Merge and split models

You can use a merge and split utility for Information Extraction models trained on different machines or with different fieldsets. The utility requires AutoML SDK version 10.2.4.6 or later.

Merge IE models

The merge functionality example is as follows:

File model1Dir = new File(getResource(this.getClass(), "util/model1").toURI());
File tempModel1Folder = Files.newTemporaryFolder();

File model2Dir = new File(getResource(this.getClass(), "util/model2").toURI());
File tempModel2Folder = Files.newTemporaryFolder();

// Merging modifies the input folders, so it is recommended to make a copy.

FileUtils.copyDirectory(model1Dir, tempModel1Folder);
FileUtils.copyDirectory(model2Dir, tempModel2Folder);

File tempModel1InputFolder = new File(tempModel1Folder.getPath() + "/training");
File tempModel2InputFolder = new File(tempModel2Folder.getPath() + "/training");

The merge result is available in tempModel1Folder:

ModelMergeUtil.merge(tempModel1InputFolder.getPath(), tempModel2InputFolder.getPath());

Split IE models

The split functionality example is as follows:

File model1Dir = new File(getResource(this.getClass(), "util/model1").toURI());
File tempModelFolder = Files.newTemporaryFolder();

// Splitting modifies the input folder, so it is recommended to make a copy.

FileUtils.copyDirectory(model1Dir, tempModelFolder);

File tempModelInputFolder = new File(tempModelFolder.getPath() + "/training");

The split results are available in the tempModelFolder and tempModelFolder_split folders created in the same location. Mind to separate features to be kept by a comma.

ModelSplitUtil.split(tempModelInputFolder.getPath(), "order_total");