In many cases, we would just want to have a script to extract the features for all samples of our database (using a specifiable ``Transformer``), so that we can use them in a different process. Currently, there is no such script available. I would propose to add a script as follows: ``` import argparse import os parser = argparse.ArgumentParser( formatter_class=argparse.ArgumentDefaultsHelpFormatter, description='Extract features from the given dataset' ) parser.add_argument("--transformer", "-e", required=True, help="Select the transformer to be used") parser.add_argument("--dataset", "-d", required=True, help="Select the dataset from which to extract features") parser.add_argument("--output-directory", "-o", required=True, help="Select the directory where to write the data to") args = parser.parse_args() import bob.bio.base import bob.core import bob.io.base logger = bob.core.log.setup("bob.paper.osijbc") bob.core.log.set_verbosity_level(logger, 2) database = bob.bio.base.load_resource(args.dataset, "database") transformer = bob.bio.base.load_resource(args.transformer, "transformer") for idx, samples in enumerate(database.all_samples()): logger.info('Extracting features for sample', ) features = transformer.transform(samples) for feature in features: output = os.path.join(args.output_directory, feature.key + ".hdf5") logger.debug('Writing file', output) bob.io.base.save(feature.data, output, True) ``` To be consistent with our other scripts, I would recommend to use `click` instead of `argparse`. Unfortunately, I am not familiar with click and I have no time to learn how to implement click commands right now. Would anyone else -- with more experience with `click` take this over?