Model builder

Description

The model builder constructs a model according to the configuration provided in the specified task and on the command line. This model can be used by MATEngine to automatically tag documents. Note that if you're using the jCarafe engine provided with MAT, the model that is built will only train to find the simple spanned annotations in documents (no spanless annotations will be trained for, and no attributes will be trained for beyond those associated with the effective label).

Note that you should never use MATModelBuilder to save models into workspaces; use MATWorkspaceEngine instead.

Note: if you create a model using this tool, and you want to do autotagging in file mode in the MAT UI, you must restart the MAT Web server. Otherwise, the UI will not be able to access the newest model.

If the annotation set corresponding to your training step contains labels which aren't intended to be trained for (e.g., they're human annotator notes which are added during hand annotation but aren't intended to be processed), be sure those labels are designated processable="no" in the annotation set descriptor.

Usage

Unix:

% $MAT_PKG_HOME/bin/MATModelBuilder

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd

Usage: MATModelBuilder [task option] [step option] [config name option] [other options]

Basic options: task, step, config name

--task <task>
Name of the task to use. Must be the first argument, if present. Obligatory if the system knows of more than one task. The system will provide a list of known tasks as part of its help string.
--step <step>
Name of the step which contains the trainable engine. Must be the first argument after --task, if present. Optional. Obligatory if the task has more than one trainable step.
--config_name <name>
Name of the model build config to use. Must be the first argument after --step, if present, or --task, if --step is not present and --task is present. Optional. Default model build config will be used if no config is specified.

Other options: input, output, etc.

--language <l>
Language to use, either by name or code, as specified in the task. Obligatory if multiple languages are present and the engine for the step in question supports multiple languages.
--input_dir <dir>
A directory, all of whose files will be used in the model construction. Can be repeated. May be specified with --input_files.
--input_files <pat>
A glob-style pattern describing full pathnames to use in the model construction. May be specified with --input_dir. Can be repeated. (If you're not familiar with Unix, glob patterns are file name patterns recognized by Unix shells. Consult your favorite Unix documentation for details.)
--file_type <t>
The file type of the input. One of the available readers. The "raw" reader is not permitted. The "mat-json" reader is the default.
--encoding <encoding>
The encoding of the input. The default is the appropriate default for the file type.
--fresh_task
If this option is present, all task information in each document will be removed and re-inferred before the model is created. Use this option if you're processing documents which were created using a task other than the current one.
--model_file <file>
Location to save the created model. The directory must already exist. Obligatory if --save_as_default_model isn't specified.
--save_as_default_model
If the the task.xml file for the task specifies the <default_model> element, save the model in the specified location, possibly overriding any existing model. The default model path receives a suffix reflecting the appropriate step and language; see here for more details.

MATModelBuilder also makes the common options available.

The reader referenced in the --file_type option may introduce additional options, which are described here. These additional options must follow the --file_type option.

The particular training engine defined for the task in your task.xml file will make available other command-line options. The command-line options for the jCarafe engine are described here. The examples below assume that you're using the jCarafe engine.

Examples

Example 1

Let's say that you have several annotated documents in /path/to/my/docs, and there are no other files in that directory. Further, you have only one task, the task has no default model, and you have a default <model_config> in your task.xml file which contains appropriate settings for the engine, feature set and PSA training. The following command would write your model to the file named "task_model" in the current directory:

Unix:

% $MAT_PKG_HOME/bin/MATModelBuilder --input_dir /path/to/my/docs --model_file $PWD/task_model

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd --input_dir c:\path\to\my\docs --model_file %CD%\task_model

To make use of this model, you could pass it to MATEngine, e.g., as the value of the --carafe_tag_model flag in our sample task.

Example 2

Let's say you have multiple tasks, and the one you want to use is "Named Entity". Your documents are in the same place, but there are other documents there too; fortunately, all the documents you want to use end with '.json'. In addition, your documents have lots of really odd person names in them, but you conveniently have a list of the names you're looking for, and you've prepared a directory /path/to/my/lexicon which contains a single file named NAMES which contains each of the tokens of interest, like so:

Urbatz
Yuguwima
Florshin
Batywan

The task you're using has a default model. The following command would save your model as the default:

Unix:

% $MAT_PKG_HOME/bin/MATModelBuilder --task "Named Entity" \
--input_files '/path/to/my/docs/*.json' \
--lexicon_dir /path/to/my/lexicon/ --save_as_default_model

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd --task "Named Entity" \
--input_files "c:\path\to\my\docs\*.json" \
--lexicon_dir c:\path\to\my\lexicon\ --save_as_default_model

Example 3

Let's say we're in the same situation as example 2, except you only want to build a model out of the files 100.json through 199.json, as well as the files in /path/to/my/other/docs.

Unix:

% $MAT_PKG_HOME/bin/MATModelBuilder --task "Named Entity" \
--input_files '/path/to/my/docs/1[0-9][0-9].json' \
--input_dir /path/to/my/other/docs \
--lexicon_dir /path/to/my/lexicon/ --save_as_default_model

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd --task "Named Entity" \
--input_files "c:\path\to\my\docs\1[0-9][0-9].json" \
--input_dir c:\path\to\my\other/docs \
--lexicon_dir c:\path\to\my\lexicon\ --save_as_default_model

Example 4

Let's say we're in the same situation as example 3, except the documents are XML inline documents with the ".xml" suffix:

Unix:

% $MAT_PKG_HOME/bin/MATModelBuilder --task "Named Entity" \
--input_files '/path/to/my/docs/1[0-9][0-9].xml' \
--file_type xml-inline \
--input_dir /path/to/my/other/docs \
--lexicon_dir /path/to/my/lexicon/ --save_as_default_model

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd --task "Named Entity" \
--input_files "c:\path\to\my\docs\1[0-9][0-9].xml" \
--file_type xml-inline \
--input_dir c:\path\to\my\other\docs \
--lexicon_dir c:\path\to\my\lexicon\ --save_as_default_model

Example 5

Let's say we're in the same situation as example 3, but we have a non-default model configuration that we want to use:

Unix:

% $MAT_PKG_HOME/bin/MATModelBuilder --task "Named Entity"
--config_name 'alt_config' \

--input_files '/path/to/my/docs/1[0-9][0-9].xml' \
--file_type xml-inline \
--input_dir /path/to/my/other/docs \
--lexicon_dir /path/to/my/lexicon/ --save_as_default_model

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd --task "Named Entity"
--config_name "alt_config" \

--input_files "c:\path\to\my\docs\1[0-9][0-9].xml" \
--file_type xml-inline \
--input_dir c:\path\to\my\other/docs \
--lexicon_dir c:\path\to\my\lexicon\ --save_as_default_model

Example 6

Let's say we're in the same situation as Example 2, but you want to build a model for the "Sample Relations" task. Since this task has multiple trainable steps, you must specify the step you're targeting:

% $MAT_PKG_HOME/bin/MATModelBuilder --task "Sample Relations" \
--step relation_tag \
--input_files '/path/to/my/docs/1[0-9][0-9].json' \
--input_dir /path/to/my/other/docs \
--lexicon_dir /path/to/my/lexicon/ --save_as_default_model

Windows native:

> %MAT_PKG_HOME%\bin\MATModelBuilder.cmd --task "Sample Relations" \
--step relation_tag \
--input_files "c:\path\to\my\docs\1[0-9][0-9].json" \
--input_dir c:\path\to\my\other/docs \
--lexicon_dir c:\path\to\my\lexicon\ --save_as_default_model