def add_active_units_args(parser: argparse.PARSER):
"""Arguments for the `active units` analysis"""
parser.add_argument(
'--mc_au_analysis',
default=0,
type=int,
help=
'number of Monte-Carlo samples used to estimate the number of active units (skip if zero)'
)
parser.add_argument('--npoints_au_analysis',
default=1000,
type=int,
help='number of data points x')
def add_iw_sweep_args(parser: argparse.PARSER, min=5, max=1e2, steps=3):
"""Arguments defining a sweep of the parameter `iw`"""
parser.add_argument('--iw_min',
default=min,
type=float,
help='min umber of Importance-Weighted samples')
parser.add_argument('--iw_max',
default=max,
type=float,
help='max number of Importance-Weighted samples')
parser.add_argument('--iw_steps',
default=steps,
type=int,
help='number of Importance-Weighted samples samples')
def configure_test_parser(subparsers: argparse.PARSER) -> None:
"""Add a testing parser to an existing argparser.
Args:
subparsers: The parser object to be appended to.
"""
parser = subparsers.add_parser('test',
description='Test a FastEstimator model',
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
allow_abbrev=False)
# use an argument group for required flag arguments since otherwise they will show up as optional in the help
parser.add_argument('entry_point', type=str, help='The path to the model python file')
parser.add_argument('--hyperparameters',
dest='hyperparameters_json',
type=str,
help="The path to the hyperparameters JSON file")
parser.add_argument_group(
'hyperparameter arguments',
'Arguments to be passed through to the get_estimator() call. \
Examples might look like --epochs <int>, --batch_size <int>, --optimizer <str>, etc...')
parser.set_defaults(func=test)
def add_gradient_analysis_args(parser: argparse.PARSER):
"""Arguments for the gradient analysis"""
parser.add_argument('--grad_bs',
default=10,
type=int,
help='gradients analysis batch size')
parser.add_argument(
'--grad_samples',
default=100,
type=int,
help=
'number of MC samples used to evaluate the Variance and SNR of the gradients.'
)
parser.add_argument(
'--grad_key',
default='inference_network',
type=str,
help=
'key matching the name of the parameters used for the gradients analysis'
)
parser.add_argument('--grad_epsilon',
default=1e-15,
type=float,
help='Minimum variance value')
from argparse import ArgumentParser, PARSER
from base64 import urlsafe_b64encode
import os
import base64
import argparse
from cryptography.fernet import Fernet
PARSER = argparse.ArgumentParser()
PARSER.add_argument_group('required arguments')
PARSER.add_argument("--token", required=True, dest="GH_TOKEN", action="store")
PROGRAM_ARGS = PARSER.parse_args()
key = Fernet.generate_key()
cipher_suite = Fernet(key)
cipher_text = cipher_suite.encrypt(bytes(PROGRAM_ARGS.GH_TOKEN, 'ascii'))
export_stmts = [
'\nexport GH_KEY="' + key.decode('utf-8') + '"\n',
'export GH_TOKEN="' + cipher_text.decode('utf-8') + '"\n'
]
env_file_path = os.environ['HOME'] + '/.zshrc'
#if MacOS, write encrypted token and key to .zshrc
try:
if os.path.exists(env_file_path):
with open(env_file_path, mode='a+') as zshrc:
zshrc.writelines(export_stmts)
zshrc.close()
except FileNotFoundError:
print("zshrc not found")
def add_model_architecture_args(parser: argparse.PARSER,
model='vae',
nlayers=3,
depth=3,
N=32):
"""Arguments defining the model architecture"""
parser.add_argument(
'--model',
default=model,
help='[vae, conv-vae, gmm, gaussian-toy, sbm, gaussian-vae]')
parser.add_argument('--hdim',
default=64,
type=int,
help='number of hidden units for each layer')
parser.add_argument('--nlayers',
default=nlayers,
type=int,
help='number of hidden layers in each MLP')
parser.add_argument(
'--depth',
default=depth,
type=int,
help='number of stochastic layers when using hierarchical models')
parser.add_argument(
'--b_nlayers',
default=1,
type=int,
help='number of MLP hidden layers for the neural baseline')
parser.add_argument(
'--normalization',
default='none',
type=str,
help=
'normalization layer for the VAE model [none | layernorm | batchnorm]')
parser.add_argument('--dropout',
default=0,
type=float,
help='dropout value')
parser.add_argument(
'--prior',
default='normal',
help='prior for the VAE model [normal, categorical, bernoulli]')
parser.add_argument(
'--N',
default=N,
type=int,
help='number of latent variables for each stochastic layer')
parser.add_argument(
'--K',
default=8,
type=int,
help='number of categories when using a categorical prior')
parser.add_argument(
'--kdim',
default=0,
type=int,
help='dimension of the keys model for categorical prior')
parser.add_argument('--learn_prior',
action='store_true',
help='learn the prior parameters (VAE model)')
def add_base_args(parser: argparse.PARSER, exp='sandbox', dataset='shapes'):
"""base experiments arguments use across all exps"""
parser.add_argument(
'--dataset',
default=dataset,
help=
'dataset [shapes | binmnist | omniglot | fashion | gmm-toy | gaussian-toy]'
)
parser.add_argument('--mini',
action='store_true',
help='use a down-sampled version of the dataset')
parser.add_argument('--root',
default='runs/',
help='directory to store training logs')
parser.add_argument('--data_root',
default='data/',
help='directory to store the data')
parser.add_argument('--exp', default=exp, help='experiment directory')
parser.add_argument('--id', default='', type=str, help='run id suffix')
parser.add_argument('--seed', default=1, type=int, help='random seed')
parser.add_argument('--workers',
default=1,
type=int,
help='number of dataloader workers')
parser.add_argument('--rf',
action='store_true',
help='delete the previous run')
parser.add_argument('--silent', action='store_true', help='silence tqdm')
parser.add_argument('--deterministic',
action='store_true',
help='use deterministic backend')
parser.add_argument(
'--sequential_computation',
action='store_true',
help='compute each iw sample sequentially during validation')
def add_run_args(parser: argparse.PARSER):
"""parse the remaining arguments for the script `run.py`"""
# epochs, batch size, MC samples, lr
parser.add_argument('--epochs',
default=-1,
type=int,
help='number of epochs (use n_steps if `epochs` < 0)')
parser.add_argument('--nsteps',
default=500000,
type=int,
help='number of optimization steps')
parser.add_argument('--optimizer',
default='adam',
help='[sgd | adam | adamax | rmsprop]')
parser.add_argument('--lr',
default=2e-3,
type=float,
help='base learning rate')
parser.add_argument('--freebits',
default=None,
type=float,
help='number of `freebits` per layer')
parser.add_argument('--baseline_lr',
default=5e-3,
type=float,
help='learning rate for the weight of the baseline')
parser.add_argument('--bs',
default=32,
type=int,
help='training batch size')
parser.add_argument('--valid_bs',
default=50,
type=int,
help='validation evaluation batch size')
parser.add_argument('--test_bs',
default=50,
type=int,
help='test evaluation batch size')
parser.add_argument('--lr_reduce_steps',
default=0,
type=int,
help='number of learning rate reduce steps')
parser.add_argument(
'--only_train_set',
action='store_true',
help=
'only use the training dataset: useful to isolate the optimization behaviour.'
)
# estimator & warmup
parser.add_argument(
'--estimator',
default='reinforce',
help=
'see estimators.config for the full list [pathwise-iwae, reinforce, reinforce-baseline, '
'ovis-S*, ovis-gamma*, tvo, vimco-arithmetic]')
parser.add_argument('--mc',
default=1,
type=int,
help='number of `outer` Monte-Carlo samples')
parser.add_argument('--iw',
default=1,
type=int,
help='number of `inner` Importance-Weighted samples')
parser.add_argument('--beta',
default=1.0,
type=float,
help='Beta weight for the KL term (i.e. Beta-VAE)')
parser.add_argument(
'--tau',
default=0,
type=float,
help=
'Temperature parameter if using the `Gumbel-Softmax` relaxation (0: `Straight-Through`)'
)
parser.add_argument(
'--alpha',
default=0,
type=float,
help='alpha weight for the unormalized weights: w_k^alpha')
parser.add_argument(
'--warmup',
default=0,
type=int,
help='period of the posterior warmup (alpha : `alpha_init` -> 0)')
parser.add_argument('--warmup_offset',
default=0,
type=int,
help='number of initial steps before increasing beta')
parser.add_argument('--warmup_mode',
default='log',
type=str,
help='interpolation mode [linear | log]')
parser.add_argument('--alpha_init',
default=0.9,
type=float,
help='initial alpha value')
# evaluation
parser.add_argument(
'--eval_freq',
default=10,
type=int,
help=
'frequency of evalution [evaluate log p(x), analyse grads and sample the prior]'
)
parser.add_argument('--max_eval',
default=None,
type=int,
help='maximum number of data points for evaluation')
parser.add_argument(
'--iw_valid',
default=100,
type=int,
help='number of Importance-Weighted samples for validation')
parser.add_argument(
'--iw_test',
default=1000,
type=int,
help='number of Importance-Weighted samples for testing')
def configure_log_parser(subparsers: argparse.PARSER) -> None:
"""Add a logging parser to an existing argparser.
Args:
subparsers: The parser object to be appended to.
"""
parser = subparsers.add_parser(
'logs',
description='Generates comparison graphs amongst one or more log files',
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
allow_abbrev=False)
parser.add_argument(
'log_dir',
metavar='<Log Dir>',
type=str,
help="The path to a folder containing one or more log files")
parser.add_argument('--extension',
metavar='E',
type=str,
help="The file type / extension of your logs",
default=".txt")
parser.add_argument(
'--recursive',
action='store_true',
help="Recursively search sub-directories for log files")
parser.add_argument(
'--ignore',
metavar='I',
type=str,
nargs='+',
help=
"The names of metrics to ignore though they may be present in the log files"
)
parser.add_argument(
'--smooth',
metavar='<float>',
type=float,
help=
"The amount of gaussian smoothing to apply (zero for no smoothing)",
default=1)
parser.add_argument('--pretty_names',
help="Clean up the metric names for display",
action='store_true')
legend_group = parser.add_argument_group('legend arguments')
legend_x_group = legend_group.add_mutually_exclusive_group(required=False)
legend_x_group.add_argument('--common_legend',
dest='share_legend',
help="Generate one legend total",
action='store_true',
default=True)
legend_x_group.add_argument('--split_legend',
dest='share_legend',
help="Generate one legend per graph",
action='store_false',
default=False)
save_group = parser.add_argument_group('output arguments')
save_x_group = save_group.add_mutually_exclusive_group(required=False)
save_x_group.add_argument(
'--save',
nargs='?',
metavar='<Save Dir>',
dest='save',
action=SaveAction,
default=False,
help="Save the output image. May be accompanied by a directory into \
which the file is saved. If no output directory is specified, the log \
directory will be used")
save_x_group.add_argument(
'--display',
dest='save',
action='store_false',
help="Render the image to the UI (rather than saving it)",
default=True)
save_x_group.set_defaults(save_dir=None)
parser.set_defaults(func=logs)