Literal & Argument Components

To supply your Xircuits library components with customizable parameters, we provide Literal and Argument components as general components. The main difference between them can be observed in the generated code. Consider the training_epoch parameter in the TrainImageClassifier component below.

You will need to supply an integer to specify how many epochs to run the training.

Literal Component

The generated code would look like this:

...
from xai_learning.training import TrainImageClassifier
from xai_learning.training import EvaluateAccuracy

def main(args):

    ...

    c_4.training_data = c_2.train
    c_4.training_epochs.value = 3 <--- Your epoch int is supplied here
    c_5.model = c_4.trained_model

    ...

    next_component = c_1.do()
    while next_component:
        next_component = next_component.do()

if __name__ == '__main__':
    parser = ArgumentParser()
    parser.add_argument('--experiment_name', default=datetime.now().strftime("%Y-%m-%d %H:%M:%S"), type=str)
    main(parser.parse_args())

As you can see, the value is passed in as a Literal Integer to the TrainImageClassifier component.

Argument Component

The following example shows if the argument component is used. Xircuits will prompt you to supply the parameter through our argument window prompt.

And it will be supplied as a argument in the generated code.

...
from xai_learning.training import TrainImageClassifier
from xai_learning.training import EvaluateAccuracy

def main(args):

    ...

    c_4.training_data = c_2.train
    c_4.training_epochs.value = args.epoch
    c_5.model = c_4.trained_model

    ...

    next_component = c_1.do()
    while next_component:
        next_component = next_component.do()

if __name__ == '__main__':
    parser = ArgumentParser()
    parser.add_argument('--experiment_name', default=datetime.now().strftime("%Y-%m-%d %H:%M:%S"), type=str)
    parser.add_argument('--epoch', default='1', type=int)
    main(parser.parse_args())

And then executed by running SampleTraining.py --epoch 3

From a user perspective, both components are perfectly executable in Xircuits, however the one supplied with arguments are more flexible if executed outside Xircuits. For example, consider the situation that you would like to automate your training workflow that displays the accuracy of your model for each training dataset. If you use Argument components, you could simply run:

python TrainingScript.py --dataset mnist