Getting Started with Domain Classifier

There are different scritps available to run the Domain Classifier. They differ in the way to manage the user interaction:

1. main_gui.py: user interacts through a Graphical User Interface (GUI).

2. main_domain_classifier.py: user interacts through a terminal (command window).

3. main_dc_single_task.py: to execute a single task (requires user interaction).

4. run_dc_task.py: to execute a single task (without user interaction).

Graphical User Interface

You will need a terminal window to launch the GUI: open a terminal, go to the roof folder of the software and run

$ python main_gui.py --p /path/to/project
                     --source /path/to/datasets
                     —-zeroshot /path/to/zeroshot

where

• /path/to/project is the path to a new or an existing project in which the application’s output will be saved.

• /path/to/datasets is the path to the source data folder.

• /path/to/zeroshot is the path to a folder containing a pre-trained zero-shot model utilized for the selection of a subcorpus from a category name.

All these parameters are optional. For instance, you can simply run

$ python main_gui.py

and select the appropriate folders inside the GUI.

Terminal mode

To use the interaction through a command window, open a terminal, go to the root folder of the software and run

$ python main_domain_classifier.py --p /path/to/project
                                   --source /path/to/datasets
                                   —-zeroshot /path/to/zeroshot

where

• /path/to/project is the path to a new or an existing project in which the application’s output will be saved.

• /path/to/datasets is the path to the source data folder.

• /path/to/zeroshot is the path to a folder containing a pre-trained zero-shot model utilized for the selection of a subcorpus from a category name.

Single task execution with user interaction

To use the interaction through a command window, open a terminal, go to the root folder of the software and run

$ python main_dc_single_task.py --p /path/to/project
                                --source /path/to/datasets
                                —-zeroshot /path/to/zeroshot
                                --task TASK

where

• /path/to/project is the path to a new or an existing project in which the application’s output will be saved.

• /path/to/datasets is the path to the source data folder.

• /path/to/zeroshot is the path to a folder containing a pre-trained zero-shot model utilized for the selection of a subcorpus from a category name.

• TASK is the name of the task to run.

You can run

$ python main_dc_single_task.py --h

to see available tasks.

Single task execution without user interaction

To run a specific task, all parameter of the task should be introduced through the comman window. To do so, you can run

$ python run_dc_task.py --p /path/to/project
                        --source /path/to/datasets
                        —-zeroshot /path/to/zeroshot
                        --task TASK
                        --class_name CLASS_NAME
                        --param1 PARAM1
                        --param1 PARAM2
                        ...

where

• /path/to/project is the path to a new or an existing project in which the application’s output will be saved.

• /path/to/datasets is the path to the source data folder.

• /path/to/zeroshot is the path to a folder containing a pre-trained zero-shot model utilized for the selection of a subcorpus from a category name.

• TASK is the name of the task to run.

• CLASS_NAME is the name of the target category (only for tasks requiring it)

• param1, param2 are the names of the specific parameters

• PARAM1, PARAM2 are their values

You can run

$ python run_dc_task.py --h

to see available tasks.