def parse_arguments(args_to_parse):
"""
Parse the command line arguments.
Params
------
args_to_parse : list of str
Arguments to parse (split on whitespaces).
"""
description = "AstraZeneca Coding Exercise"
default_config = get_config_section([CONFIG_FILE], "Preset")
parser = argparse.ArgumentParser(description=description,
formatter_class=FormatterNoDuplicate)
# API options
api = parser.add_argument_group('API specific options')
api.add_argument('-u',
'--base-url',
help="Base url for API query",
default=default_config['base_url'])
api.add_argument('-m',
'--manufacturer',
help='Drug manufacturer',
default=default_config['manufacturer'])
# Plotting options
plots = parser.add_argument_group('Plot specific options')
plots.add_argument('-p',
'--plot-type',
default=default_config['plot_type'],
choices=PLOTS,
help='Type of plot to produce')
plots.add_argument('-s',
'--save-fig',
help='Whether to save the figure',
type=bool,
default=default_config['save_fig'])
# Analysis options
analysis_options = parser.add_argument_group('Anaylsis options')
analysis_options.add_argument('-a',
'--analysis-type',
type=str,
choices=ANALYSIS_TYPE,
default=default_config['analysis_type'],
help="The type of analysis to undertake")
args = parser.parse_args(args_to_parse)
return args
def parse_arguments(args_to_parse):
"""Parse the command line arguments.
Parameters
----------
args_to_parse: list of str
Arguments to parse (splitted on whitespaces).
"""
default_config = get_config_section([CONFIG_FILE], "Custom")
description = "PyTorch implementation and evaluation of disentangled Variational AutoEncoders and metrics."
parser = argparse.ArgumentParser(description=description,
formatter_class=FormatterNoDuplicate)
# General options
general = parser.add_argument_group('General options')
general.add_argument(
'name',
type=str,
help="Name of the model for storing and loading purposes.")
general.add_argument('-L',
'--log-level',
help="Logging levels.",
default=default_config['log_level'],
choices=LOG_LEVELS)
general.add_argument('--no-progress-bar',
action='store_true',
default=default_config['no_progress_bar'],
help='Disables progress bar.')
general.add_argument('--no-cuda',
action='store_true',
default=default_config['no_cuda'],
help='Disables CUDA training, even when have one.')
general.add_argument(
'-s',
'--seed',
type=int,
default=default_config['seed'],
help='Random seed. Can be `None` for stochastic behavior.')
# Learning options
training = parser.add_argument_group('Training specific options')
training.add_argument(
'--checkpoint-every',
type=int,
default=default_config['checkpoint_every'],
help='Save a checkpoint of the trained model every n epoch.')
training.add_argument('-d',
'--dataset',
help="Path to training data.",
default=default_config['dataset'],
choices=DATASETS)
training.add_argument(
'-x',
'--experiment',
default=default_config['experiment'],
choices=EXPERIMENTS,
help=
'Predefined experiments to run. If not `custom` this will overwrite some other arguments.'
)
training.add_argument('-e',
'--epochs',
type=int,
default=default_config['epochs'],
help='Maximum number of epochs to run for.')
training.add_argument('-b',
'--batch-size',
type=int,
default=default_config['batch_size'],
help='Batch size for training.')
training.add_argument('--lr',
type=float,
default=default_config['lr'],
help='Learning rate.')
# Model Options
model = parser.add_argument_group('Model specfic options')
model.add_argument('-m',
'--model-type',
default=default_config['model'],
choices=MODELS,
help='Type of encoder and decoder to use.')
model.add_argument('-z',
'--latent-dim',
type=int,
default=default_config['latent_dim'],
help='Dimension of the latent variable.')
model.add_argument('-l',
'--loss',
default=default_config['loss'],
choices=LOSSES,
help="Type of VAE loss function to use.")
model.add_argument('-r',
'--rec-dist',
default=default_config['rec_dist'],
choices=RECON_DIST,
help="Form of the likelihood ot use for each pixel.")
model.add_argument(
'-a',
'--reg-anneal',
type=float,
default=default_config['reg_anneal'],
help=
"Number of annealing steps where gradually adding the regularisation. What is annealed is specific to each loss."
)
# Loss Specific Options
betaH = parser.add_argument_group('BetaH specific parameters')
betaH.add_argument('--betaH-B',
type=float,
default=default_config['betaH_B'],
help="Weight of the KL (beta in the paper).")
betaB = parser.add_argument_group('BetaB specific parameters')
betaB.add_argument('--betaB-initC',
type=float,
default=default_config['betaB_initC'],
help="Starting annealed capacity.")
betaB.add_argument('--betaB-finC',
type=float,
default=default_config['betaB_finC'],
help="Final annealed capacity.")
betaB.add_argument(
'--betaB-G',
type=float,
default=default_config['betaB_G'],
help="Weight of the KL divergence term (gamma in the paper).")
factor = parser.add_argument_group('factor VAE specific parameters')
factor.add_argument('--factor-G',
type=float,
default=default_config['factor_G'],
help="Weight of the TC term (gamma in the paper).")
factor.add_argument('--lr-disc',
type=float,
default=default_config['lr_disc'],
help='Learning rate of the discriminator.')
btcvae = parser.add_argument_group('beta-tcvae specific parameters')
btcvae.add_argument('--btcvae-A',
type=float,
default=default_config['btcvae_A'],
help="Weight of the MI term (alpha in the paper).")
btcvae.add_argument(
'--btcvae-G',
type=float,
default=default_config['btcvae_G'],
help="Weight of the dim-wise KL term (gamma in the paper).")
btcvae.add_argument('--btcvae-B',
type=float,
default=default_config['btcvae_B'],
help="Weight of the TC term (beta in the paper).")
# Learning options
evaluation = parser.add_argument_group('Evaluation specific options')
evaluation.add_argument(
'--is-eval-only',
action='store_true',
default=default_config['is_eval_only'],
help='Whether to only evaluate using precomputed model `name`.')
evaluation.add_argument(
'--is-metrics',
action='store_true',
default=default_config['is_metrics'],
help=
"Whether to compute the disentangled metrcics. Currently only possible with `dsprites` as it is the only dataset with known true factors of variations."
)
evaluation.add_argument('--no-test',
action='store_true',
default=default_config['no_test'],
help="Whether not to compute the test losses.`")
evaluation.add_argument('--eval-batchsize',
type=int,
default=default_config['eval_batchsize'],
help='Batch size for evaluation.')
args = parser.parse_args(args_to_parse)
if args.experiment != 'custom':
if args.experiment not in ADDITIONAL_EXP:
# update all common sections first
model, dataset = args.experiment.split("_")
common_data = get_config_section([CONFIG_FILE],
"Common_{}".format(dataset))
update_namespace_(args, common_data)
common_model = get_config_section([CONFIG_FILE],
"Common_{}".format(model))
update_namespace_(args, common_model)
try:
experiments_config = get_config_section([CONFIG_FILE],
args.experiment)
update_namespace_(args, experiments_config)
except KeyError as e:
if args.experiment in ADDITIONAL_EXP:
raise e # only reraise if didn't use common section
return args