def add_model_specific_args(parent_parser):
p = HyperOptArgumentParser(parents=parent_parser, add_help=False, conflict_handler='resolve')
p.add_argument('--dense_encoder', default=False, type=bool)
p.add_argument('--code_size', default=512, type=int)
p.add_argument('--out_channels', default=3, type=int)
p.add_argument('--rec_decoder_convl', default=3, type=int)
p.add_argument('--comp_decoder_convl', default=3, type=int)
p.add_argument('--context_decoder_convl', default=3, type=int)
if not parent_parser: # Name clash with parent
p.add_argument('--in_channels', default=3, type=int)
return p
def add_model_specific_args(parent_parser, root_dir): # pragma: no cover
"""
Parameters you define here will be available to your model through self.hparams
:param parent_parser:
:param root_dir:
:return:
"""
parser = HyperOptArgumentParser(
strategy=parent_parser.strategy, parents=[parent_parser]
)
parser.add_argument("--beta", default=1, type=float)
parser.add_argument("--bottleneck_size", default=10, type=int)
parser.add_argument("--group_distance_scaling", default=1, type=float)
parser.add_argument("--audio_size", default=80, type=int)
parser.add_argument("--data_dim", default=140, type=int)
parser.add_argument("--data_files", default="datafiles.json", type=str)
parser.add_argument("--data_dir", default="/data_dir", type=str)
parser.opt_list(
"--batch_size",
default=256 * 4,
type=int,
options=[32, 64, 128, 256],
tunable=False,
help="batch size will be divided over all gpus being used across all nodes",
)
parser.opt_list(
"--learning_rate",
default=0.001 * 8,
type=float,
options=[0.0001, 0.0005, 0.001],
tunable=True,
)
return parser
default=10,
options=[20, 30, 40, 50],
type=int,
tunable=True)
parser.add_opt_argument_list('--embedding_dim',
default=320,
options=[100, 200, 320, 400],
type=int,
tunable=True)
parser.add_opt_argument_list('--max_seq_len',
default=50,
options=[50, 70, 90, 110],
type=int,
tunable=True)
parser.add_argument('--nb_epochs', default=5, type=int)
parser.add_argument('--optimizer_name', default='adam')
parser.add_argument('--eval_every_n_batches', default=200, type=int)
parser.add_argument('--train_mode', default='train')
parser.add_argument('--nb_grams', default=2, type=int)
parser.add_argument('--root_dir', default='/home/ankur/smart-reply')
parser.add_argument(
'--dataset_train_path',
default='/home/ankur/smart-reply/dataset/data/data/train.csv')
parser.add_argument(
'--dataset_test_path',
default='/home/ankur/smart-reply/dataset/data/data/test.csv')
parser.add_argument(
'--dataset_val_path',
]
checkpoint_path = experiment_folder + "/checkpoints/" + checkpoints[-1]
model = BERTClassifier.load_from_metrics(weights_path=checkpoint_path,
tags_csv=tags_csv_file)
# Make sure model is in prediction mode
model.eval()
model.freeze()
return model
if __name__ == "__main__":
parser = HyperOptArgumentParser(description="Minimalist BERT Classifier",
add_help=True)
parser.add_argument(
"--experiment",
required=True,
type=str,
help="Path to the experiment folder.",
)
hparams = parser.parse_args()
print("Loading model...")
model = load_model_from_experiment(hparams.experiment)
print(model)
while 1:
print(
"Please write a movie review or quit to exit the interactive shell:"
)
# Get input sentence
input_sentence = input("> ")
if input_sentence == "q" or input_sentence == "quit":
break
if __name__ == '__main__':
# dirs
root_dir = os.path.dirname(os.path.realpath(__file__))
demo_log_dir = os.path.join(root_dir, 'pt_lightning_demo_logs')
checkpoint_dir = os.path.join(demo_log_dir, 'model_weights')
test_tube_dir = os.path.join(demo_log_dir, 'test_tube_data')
# although we user hyperOptParser, we are using it only as argparse right now
parent_parser = HyperOptArgumentParser(strategy='grid_search',
add_help=False)
# gpu args
parent_parser.add_argument('--gpus',
type=str,
default='-1',
help='how many gpus to use in the node.'
' value -1 uses all the gpus on the node')
parent_parser.add_argument('--test_tube_save_path',
type=str,
default=test_tube_dir,
help='where to save logs')
parent_parser.add_argument('--model_save_path',
type=str,
default=checkpoint_dir,
help='where to save model')
parent_parser.add_argument('--experiment_name',
type=str,
default='pt_lightning_exp_a',
help='test tube exp name')
def add_model_specific_args(parent_parser):
parser = HyperOptArgumentParser(parents=[parent_parser])
parser.add_argument('--num_freq_mask', default=2, type=int)
parser.add_argument('--num_time_mask', default=2, type=int)
parser.add_argument('--freq_mask_length', default=20, type=int)
parser.add_argument('--time_mask_length', default=20, type=int)
parser.add_argument('--feature_dim', default=480, type=int)
parser.add_argument('--model_size', default=512, type=int)
parser.add_argument('--feed_forward_size', default=2048, type=int)
parser.add_argument('--hidden_size', default=64, type=int)
parser.add_argument('--dropout', default=0.1, type=float)
parser.add_argument('--num_head', default=8, type=int)
parser.add_argument('--num_encoder_layer', default=12, type=int)
parser.add_argument('--num_decoder_layer', default=6, type=int)
parser.add_argument('--vocab_path',
default='testing_vocab.model',
type=str)
parser.add_argument('--max_feature_length', default=1024, type=int)
parser.add_argument('--max_token_length', default=70, type=int)
parser.add_argument('--share_weight', default=True, type=bool)
parser.add_argument('--loss_lambda', default=0.65, type=float)
parser.add_argument('--smoothing', default=0.1, type=float)
parser.add_argument('--use_low_rank', default=False, type=bool)
parser.add_argument('--lr', default=5e-5, type=float)
parser.add_argument('--warm_up_step', default=16000, type=int)
parser.add_argument('--factor', default=1, type=int)
parser.add_argument('--enable_spec_augment', default=True, type=bool)
parser.add_argument('--train_batch_size', default=64, type=int)
parser.add_argument('--train_loader_num_workers', default=32, type=int)
parser.add_argument('--val_batch_size', default=64, type=int)
parser.add_argument('--val_loader_num_workers', default=16, type=int)
return parser
# 5 START TRAINING
# ------------------------
trainer.fit(model)
if __name__ == "__main__":
# ------------------------
# TRAINING ARGUMENTS
# ------------------------
# these are project-wide arguments
parser = HyperOptArgumentParser(
strategy="random_search",
description="Minimalist BERT Classifier",
add_help=True,
)
parser.add_argument("--seed", type=int, default=3, help="Training seed.")
parser.add_argument(
"--save_top_k",
default=1,
type=int,
help=
"The best k models according to the quantity monitored will be saved.",
)
# Early Stopping
parser.add_argument("--monitor",
default="val_acc",
type=str,
help="Quantity to monitor.")
parser.add_argument(
"--metric_mode",
default="max",
weight_decay_rate=hparams.wd)
print(hparams, flush=True)
# start trainer
auc = train(vars(hparams), num_workers, node_id, fold, kfold, debug, path)
print(auc)
# save Experiment
exp.add_scalar('auc', auc)
exp.save()
if __name__ == '__main__':
# Set up our argparser and make the y_val tunable.
parser = HyperOptArgumentParser(strategy='random_search')
parser.add_argument('--test_tube_exp_name', default='tt_exp')
parser.add_argument('--batch_size', default=80, type=int)
parser.add_argument('--epochs', default=3, type=int)
parser.add_argument('--gpus', default=2, type=int)
parser.add_argument('--log_path', default=os.environ['SCRATCH'] + '/summer_school/hs_log')
parser.opt_range('--lr', default=2e-4, type=float, tunable=True, low=1e-4, high=1e-3, nb_samples=100, log_base=10)
parser.opt_range('--wd', default=1e-5, type=float, tunable=True, low=1e-7, high=1e-4, nb_samples=100, log_base=10)
hyperparams = parser.parse_args()
# Enable cluster training.
cluster = SlurmCluster(
hyperparam_optimizer=hyperparams,
log_path=hyperparams.log_path,
python_cmd='python3',
)
tunable=False,
low=0.,
high=.5,
help='dropout fraction',
default=0.)
parser.opt_list('--batch_norm',
action='store_true',
dest='batch_norm',
tunable=False,
options=[True, False],
help='batch normalization',
default=False)
parser.add_argument('--num_accumulate',
action='store',
dest='num_accumulate',
type=int,
help='nunumber of batch accumulations',
default=1)
parser.add_argument('--name',
action='store',
dest='name',
type=str,
help='experiment name',
default=1)
parser.add_argument('--version',
action='store',
dest='version',
type=int,
help='version number',
default=None)
parser.add_argument('--gpu',
def cnn_encoder_decoder_argparser():
# TODO: set min max size for vocab
# TODO: specify optimizer
# TODO: set max length of sentences
# parser = argparse.ArgumentParser(description='Run Tests')
parser = HyperOptArgumentParser(description='Run Tests',
strategy='grid_search')
parser.add_argument('--name',
help="experiment name",
default="rnn_encoder_decoder")
parser.add_argument('--test',
help="Run test instead of training/validation",
action="store_true")
parser.add_argument("--njobs",
help="Number of jobs to use when loading data",
type=int,
default=1)
parser.add_argument("--epoch",
help="Number of epoch to run",
type=int,
default=10000)
parser.add_argument("--batch_size",
help="Batch size",
type=int,
default=32)
parser.add_argument("--hidden_size",
help="Hidden size",
type=int,
default=256)
parser.add_argument("--embedding_size",
help="Embedding size",
type=int,
default=256)
parser.add_argument("--print_every",
help="How frequently print result",
type=int,
default=100)
parser.add_argument("--lr",
help="Learning Rate",
type=float,
default=1e-03)
parser.add_argument("--dropout",
help="Dropout Rate",
type=float,
default=0)
parser.add_argument(
'--bidirectional',
help="Use bidirectional RNN in encoder",
action="store_true") # don't set this true in this model
parser.add_argument('--cpu',
help="Use cpu instead of gpu",
action="store_true")
parser.add_argument("--data",
help="Directory where data is stored",
default='../data/iwslt-zh-en/')
parser.add_argument("--rnn_type",
help="Which rnn to use (rnn, lstm, gru)",
default='gru')
parser.add_argument("--kernel_size",
help="Kernel size of cnn layers in encoder",
type=int,
default=3)
parser.add_argument("--num_encoder_layers",
help="Number of cnn layers in encoder",
type=int,
default=1)
parser.add_argument("--num_decoder_layers",
help="Number of rnn layers in encoder",
type=int,
default=1)
parser.add_argument("--early_stopping",
help="Stop if validation does not improve",
type=int,
default=10)
parser.add_argument('--l2_penalty',
help="L2 pelnalty coefficient in optimizer",
type=float,
default=0) #1e-06
parser.add_argument('--clip',
help="clip coefficient in optimizer",
type=float,
default=1)
parser.add_argument('--teacher_forcing',
help="probability of performing teacher forcing",
type=float,
default=0.5)
parser.add_argument("--max_sentence_length",
help="maximum sentence length",
type=int,
default=50)
parser.add_argument('--split_chinese_into_characters',
help="Split chinese into characters",
action="store_true")
parser.add_argument(
"--min_freq",
help="Vocabulary needs to be present at least this amount of time",
type=int,
default=3)
parser.add_argument("--max_vocab_size",
help="At most n vocaburaries are kept in the model",
type=int,
default=100000)
parser.add_argument("--source_lang",
help="Source language (vi, zh)",
default="zh")
parser.add_argument("--logs_path",
help="Path to save training logs",
type=str,
default="../training_logs/")
parser.add_argument("--model_weights_path",
help="Path to save best model weights",
type=str,
default="../model_weights/")
parser.add_argument('--save_all_epoch',
help="save all epoch",
action="store_true")
parser.add_argument('--relu',
help="use relu in decoder after embedding",
action="store_true")
parser.add_argument('--attention',
help="use luong attention decoder",
action="store_true")
parser.add_argument("--attn_model",
help="dot, general, or concat",
default='general')
parser.add_argument("--preprocess_version",
help="Version of preprocessing",
type=int,
default=2)
parser.add_argument(
'--freeze_all_words',
help="freeze word embedding and use character embedding from ELMo",
action="store_true")
parser.add_argument("--encoder_word_embedding",
help="Word embedding weights file",
default=None)
parser.add_argument("--decoder_word_embedding",
help="Word embedding weights file",
default=None)
return parser
def add_model_specific_args(parent_parser):
from test_tube import HyperOptArgumentParser
parser = HyperOptArgumentParser(parents=[parent_parser],
add_help=False)
parser.add_argument('--base_encoder', type=str, default='resnet18')
parser.add_argument('--emb_dim', type=int, default=128)
parser.add_argument('--num_workers', type=int, default=8)
parser.add_argument('--num_negatives', type=int, default=65536)
parser.add_argument('--encoder_momentum', type=float, default=0.999)
parser.add_argument('--softmax_temperature', type=float, default=0.07)
parser.add_argument('--optimizer', type=str, default='SGD')
parser.add_argument('--optimizer-parameters',
type=json.loads,
default=json.dumps({
"lr": 0.03,
"momentum": 0.9,
"weight_decay": 1.0e-4
}))
# parser.add_argument('--learning_rate', type=float, default=0.03)
# parser.add_argument('--momentum', type=float, default=0.9)
# parser.add_argument('--weight_decay', type=float, default=1e-4)
parser.add_argument('--data_dir', type=str, default='./')
parser.add_argument('--dataset',
type=str,
default='cifar10',
help='cifar10, stl10, imagenet2012')
parser.add_argument('--batch_size', type=int, default=256)
parser.add_argument('--use_mlp', action='store_true')
parser.add_argument('--meta_root',
default='.',
type=str,
help='path to meta.bin for imagenet')
return parser
def add_model_specific_args(parent_parser, root_dir):
parser = HyperOptArgumentParser(strategy=parent_parser.strategy,
parents=[parent_parser])
parser.set_defaults(nb_hopt_trials=1000)
parser.set_defaults(min_nb_epochs=1000)
parser.set_defaults(max_nb_epochs=1100)
parser.set_defaults(early_stop_metric='val_nce')
parser.set_defaults(model_save_monitor_value='val_nce')
parser.set_defaults(model_save_monitor_mode='min')
parser.set_defaults(early_stop_mode='min')
# CIFAR 10
dataset_name = 'CIFAR10'
image_height = 32
nb_classes = 10
patch_size = 8
patch_overlap = 4
# dataset options
parser.opt_list('--nb_classes',
default=nb_classes,
type=int,
options=[10],
tunable=False)
parser.opt_list('--patch_size',
default=patch_size,
type=int,
options=[10],
tunable=False)
parser.opt_list('--patch_overlap',
default=patch_overlap,
type=int,
options=[10],
tunable=False)
# network params
parser.add_argument('--image_height', type=int, default=image_height)
# trainin params
parser.add_argument('--dataset_name', type=str, default=dataset_name)
parser.add_argument('--batch_size',
type=int,
default=200,
help='input batch size (default: 200)')
parser.opt_list(
'--learning_rate',
type=float,
default=0.0002,
options=[
2e-4 * (1 / 64),
2e-4 * (1 / 32),
2e-4 * (1 / 16),
2e-4 * (1 / 8),
2e-4 * (1 / 4),
2e-4 * (1 / 2),
2e-4 * (1 / 4),
2e-4,
2e-4 * 4, #2e-4*4,
2e-4 * 8,
],
tunable=False)
# data
parser.opt_list('--cifar10_root',
default=f'{root_dir}/fisherman/datasets',
type=str,
tunable=False)
return parser
def add_model_specific_args(parent_parser, root_dir): # pragma: no cover
"""
Parameters you define here will be available to your model through self.hparams
:param parent_parser:
:param root_dir:
:return:
"""
parser = HyperOptArgumentParser(strategy=parent_parser.strategy,
parents=[parent_parser])
parser.add_argument('--log_dir', default='logs', type=str)
parser.add_argument('--checkpoint_dir',
default='checkpoints',
type=str)
# data
parser.add_argument('--training_set_path',
default='data/training',
type=str)
parser.add_argument('--validation_set_path',
default='data/validation',
type=str)
parser.add_argument('--test_task_set_path',
default='data/test_task',
type=str)
parser.add_argument('--unique_steps', default=8., type=float)
parser.add_argument('--sampling_rate', default=44100, type=float)
parser.add_argument('--dummy_datasets', default=False, type=bool)
# preprocessing
parser.add_argument('--cqt_fmin', default=30., type=float)
parser.add_argument('--cqt_n_bins', default=292, type=int)
parser.add_argument('--cqt_bins_per_octave', default=32, type=int)
parser.add_argument('--cqt_hop_length', default=256, type=int)
parser.add_argument('--cqt_filter_scale', default=0.5, type=float)
parser.add_argument('--scalogram_output_power', default=1., type=float)
parser.add_argument('--scalogram_pooling',
nargs='+',
default=(1, 2),
type=int)
parser.add_argument('--phase', default=True, type=bool)
parser.add_argument('--audio_noise', default=0., type=float)
# training
parser.opt_list('--optimizer_name',
default='adam',
type=str,
options=['adam'],
tunable=False)
parser.opt_list('--learning_rate',
default=0.0001,
type=float,
options=[0.0001, 0.0005, 0.001],
tunable=True)
parser.add_argument('--batch_size', default=32, type=int)
parser.add_argument('--score_over_all_timesteps',
default=True,
type=bool)
parser.add_argument('--visible_steps', default=60, type=int)
parser.add_argument('--prediction_steps', default=16, type=int)
parser.add_argument('--wasserstein_penalty', default=0., type=float)
parser.add_argument('--detach_targets', default=False, type=bool)
parser.add_argument('--file_batch_size', default=1, type=int)
parser.add_argument('--use_all_GPUs', default=True, type=bool)
parser.add_argument('--warmup_steps', default=0, type=int)
parser.add_argument('--annealing_steps', default=1e6, type=int)
# encoder
parser.add_argument('--enc_channels',
nargs='+',
default=(1, 32, 32, 64, 64, 128, 128, 256, 512),
type=int)
parser.add_argument('--enc_kernel_1_w',
nargs='+',
default=(1, 3, 1, 3, 1, 3, 1, 3),
type=int)
parser.add_argument('--enc_kernel_1_h',
nargs='+',
default=(65, 3, 33, 3, 16, 3, 9, 3),
type=int)
parser.add_argument('--enc_kernel_2_w',
nargs='+',
default=(3, 3, 3, 3, 3, 3, 3, 3),
type=int)
parser.add_argument('--enc_kernel_2_h',
nargs='+',
default=(3, 3, 3, 3, 3, 3, 3, 3),
type=int)
parser.add_argument('--enc_padding_1',
nargs='+',
default=(0, 1, 0, 1, 0, 1, 0, 0),
type=int)
parser.add_argument('--enc_padding_2',
nargs='+',
default=(1, 1, 1, 1, 1, 1, 1, 0),
type=int)
parser.add_argument('--enc_pooling_1',
nargs='+',
default=(1, 1, 1, 1, 1, 1, 1, 1),
type=int)
parser.add_argument('--enc_pooling_2',
nargs='+',
default=(1, 1, 1, 1, 1, 1, 1, 1),
type=int)
parser.add_argument('--enc_stride_1',
nargs='+',
default=(1, 1, 1, 1, 1, 1, 1, 1),
type=int)
parser.add_argument('--enc_stride_2',
nargs='+',
default=(2, 1, 2, 1, 2, 1, 1, 1),
type=int)
parser.add_argument('--enc_batch_norm', default=True, type=bool)
parser.add_argument('--enc_batch_norm_affine',
default=False,
type=bool)
parser.add_argument('--enc_dropout', default=0.2, type=float)
parser.add_argument('--enc_residual', default=True, type=bool)
parser.add_argument('--enc_bias', default=True, type=bool)
# ar model
parser.add_argument('--ar_channels',
nargs='+',
default=(512, 512, 512, 512, 512, 256, 256, 256,
256, 256),
type=int)
parser.add_argument('--ar_kernel_sizes',
nargs='+',
default=(5, 4, 1, 3, 3, 1, 3, 1, 5),
type=int)
parser.add_argument('--ar_pooling',
nargs='+',
default=(1, 1, 2, 1, 1, 2, 1, 1, 1),
type=int)
parser.add_argument('--ar_stride',
nargs='+',
default=(1, 1, 1, 1, 1, 1, 1, 1, 1),
type=int)
parser.add_argument('--ar_self_attention',
nargs='+',
default=(False, True, False, False, True, False,
True, False, False),
type=bool)
parser.add_argument('--ar_batch_norm', default=True, type=bool)
parser.add_argument('--ar_batch_norm_affine', default=False, type=bool)
parser.add_argument('--ar_dropout', default=0.2, type=float)
parser.add_argument('--ar_residual', default=True, type=bool)
parser.add_argument('--ar_bias', default=True, type=bool)
return parser
torch.FloatTensor(content['wordEmbeddings']).to(device)
}
print('Training div: {} BLength {} {}'.format(
train_div, p2r_data['bridge_length'],
len(p2r_data['bridge_ids'].tolist())))
return p2r_data
if __name__ == '__main__':
# use default args given by lightning
root_dir = os.path.dirname(
os.path.realpath(sys.modules['__main__'].__file__))
parent_parser = HyperOptArgumentParser(strategy='grid_search',
add_help=False)
parent_parser.add_argument('--root_dir', default=root_dir, type=str)
parent_parser.add_argument('--exp_name',
default='p2r_experiment',
type=str)
# allow model to overwrite or extend args
parser = P2rSystem.add_model_specific_args(parent_parser, root_dir)
hparams = parser.parse_args()
p2r_data = loadData(hparams.data_root, hparams.train_div)
# train model
# for trial_hparams in hparams.trials(2):
main(hparams, p2r_data)
def get_args():
parser = HyperOptArgumentParser()
parser.add_argument('--work_dir', metavar='DIR', default="./work_dir", type=str, help='path to save output')
parser.add_argument('--name', type=str)
parser.add_argument('--gpus', type=str, default='7', help='how many gpus')
parser.add_argument('--dist_bd', type=str, default='dp', choices=('dp', 'ddp', 'ddp2'),
help='supports three options dp, ddp, ddp2')
parser.add_argument('--use_16bit', dest='use_16bit', action='store_true',
help='if true uses 16 bit precision')
parser.add_argument('--eval', '--evaluate', dest='evaluate', action='store_true',
help='evaluate model on validation set')
parser.add_argument('--seed', default=1, type=int)
parser.add_argument('--load_mem', action='store_true')
parser.add_argument('--track_grad_norm', action='store_true')
parser = CGCNModel.add_model_specific_args(parser)
return parser.parse_args()
parser = HyperOptArgumentParser(usage=usage_str, description=description_str, formatter_class=argparse.ArgumentDefaultsHelpFormatter, strategy='random_search')
parser.opt_range('--step', type=float, dest='step', default=.001, help='step size/learning rate', tunable=True, nb_samples=100, low=.0001, high=.001)
parser.opt_range('--l2lam_init', action='store', type=float, dest='l2lam_init', default=.001, tunable=False, low=.0001, high=100, help='initial l2 regularization')
parser.opt_list('--solver', action='store', dest='solver', type=str, tunable=False, options=['sgd', 'adam'], help='optimizer/solver ("adam", "sgd")', default="adam")
parser.opt_range('--cg_max_iter', action='store', dest='cg_max_iter', type=int, tunable=False, low=1, high=20, help='max number of conjgrad iterations', default=10)
parser.opt_range('--batch_size', action='store', dest='batch_size', type=int, tunable=False, low=1, high=20, help='batch size', default=2)
parser.opt_range('--num_unrolls', action='store', dest='num_unrolls', type=int, tunable=False, low=1, high=10, nb_samples=4, help='number of unrolls', default=4)
parser.opt_range('--num_admm', action='store', dest='num_admm', type=int, tunable=False, low=1, high=10, nb_samples=4, help='number of ADMM iterations', default=3)
parser.opt_list('--network', action='store', dest='network', type=str, tunable=False, options=['ResNet', 'ResNet5Block'], help='which denoiser network to use', default='ResNet')
parser.opt_list('--latent_channels', action='store', dest='latent_channels', type=int, tunable=False, options=[16, 32, 64, 128], help='number of latent channels', default=64)
parser.opt_range('--num_blocks', action='store', dest='num_blocks', type=int, tunable=False, low=1, high=4, nb_samples=3, help='number of ResNetBlocks', default=3)
parser.opt_range('--dropout', action='store', dest='dropout', type=float, tunable=False, low=0., high=.5, help='dropout fraction', default=0.)
parser.opt_list('--batch_norm', action='store_true', dest='batch_norm', tunable=False, options=[True, False], help='batch normalization', default=False)
parser.add_argument('--num_accumulate', action='store', dest='num_accumulate', type=int, help='nunumber of batch accumulations', default=1)
parser.add_argument('--name', action='store', dest='name', type=str, help='experiment name', default=1)
parser.add_argument('--version', action='store', dest='version', type=int, help='version number', default=None)
parser.add_argument('--gpu', action='store', dest='gpu', type=str, help='gpu number(s)', default=None)
parser.add_argument('--num_epochs', action='store', dest='num_epochs', type=int, help='number of epochs', default=20)
parser.add_argument('--random_seed', action='store', dest='random_seed', type=int, help='random number seed for numpy', default=723)
parser.add_argument('--recon', action='store', type=str, dest='recon', default='cgsense', help='reconstruction method')
parser.add_argument('--data_file', action='store', dest='data_file', type=str, help='data.h5', default=None)
parser.add_argument('--data_val_file', action='store', dest='data_val_file', type=str, help='data.h5', default=None)
parser.add_argument('--num_data_sets', action='store', dest='num_data_sets', type=int, help='number of data sets to use', default=None)
parser.add_argument('--num_workers', action='store', type=int, dest='num_workers', help='number of workers', default=0)
parser.add_argument('--shuffle', action='store_true', dest='shuffle', help='shuffle input data files each epoch', default=False)
parser.add_argument('--clip_grads', action='store', type=float, dest='clip_grads', help='clip norm of gradient vector to val', default=0)
parser.add_argument('--cg_eps', action='store', type=float, dest='cg_eps', help='conjgrad eps', default=1e-4)
parser.add_argument('--stdev', action='store', type=float, dest='stdev', help='complex valued noise standard deviation', default=0.)
parser.add_argument('--max_norm_constraint', action='store', type=float, dest='max_norm_constraint', help='norm constraint on weights', default=None)
# define tensorflow graph
x = tf.placeholder(dtype=tf.int32, name='x')
y = tf.placeholder(dtype=tf.int32, name='y')
out = x * y
sess = tf.Session()
# Run the tf op
for train_step in range(0, 100):
output = sess.run(out, feed_dict={x: hparams.x_val, y: hparams.y_val})
exp.log({'fake_err': output})
# save exp when we're done
exp.save()
# set up our argparser and make the y_val tunable
parser = HyperOptArgumentParser(strategy='random_search')
parser.add_argument('--path', default='some/path')
parser.opt_list('--y_val', default=12, options=[1, 2, 3, 4], tunable=True)
parser.opt_list('--x_val', default=12, options=[20, 12, 30, 45], tunable=True)
hyperparams = parser.parse_args()
# optimize on 4 gpus at the same time
# each gpu will get 1 experiment with a set of hyperparams
hyperparams.optimize_parallel_gpu(train,
gpu_ids=['1', '0', '3', '2'],
nb_trials=4,
nb_workers=4)
if __name__ == '__main__':
# use default args
root_dir = os.path.dirname(os.path.realpath(__file__))
demo_log_dir = os.path.join(root_dir, 'pt_lightning_demo_logs')
checkpoint_dir = os.path.join(demo_log_dir, 'model_weights')
test_tube_dir = os.path.join(demo_log_dir, 'test_tube_data')
slurm_out_dir = os.path.join(demo_log_dir, 'slurm_scripts')
parent_parser = HyperOptArgumentParser(strategy='grid_search', add_help=False)
# cluster args not defined inside the model
parent_parser.add_argument('--per_experiment_nb_gpus', type=int,
default=8, help='how many gpus to use in a node')
parent_parser.add_argument('--nb_gpu_nodes', type=int, default=2,
help='how many nodes to use in a cluster')
parent_parser.add_argument('--test_tube_save_path', type=str, default=test_tube_dir,
help='where to save logs')
parent_parser.add_argument('--slurm_log_path', type=str, default=slurm_out_dir,
help='where to save slurm meta')
parent_parser.add_argument('--model_save_path', type=str, default=checkpoint_dir,
help='where to save model')
parent_parser.add_argument('--distributed_backend', type=str, default='ddp',
help='ddp or ddp2')
parent_parser.add_argument('--experiment_name', type=str, default='pt_lightning_exp_a',
help='test tube exp name')
parent_parser.add_argument('--num_hyperparam_trials', type=int, default=6,
help='how many grid search trials to run')
def _train(_hparams):
model = CNN(_hparams)
trainer = Trainer(
logger=TestTubeLogger(save_dir=".",
name="logs_{}".format(_hparams.name)),
gpus=_hparams.gpus,
row_log_interval=10,
log_save_interval=100,
# fast_dev_run=True,
# overfit_pct=0.01,
early_stop_callback=EarlyStopping(patience=3),
)
trainer.fit(model)
trainer.test()
if __name__ == '__main__':
parser = HyperOptArgumentParser(strategy='grid_search', add_help=False)
parser.add_argument('--gpus', type=str, default=None)
parser.add_argument('--train', action='store_true')
parser.add_argument('--name', default='CNN', type=str)
parser = CNN.add_model_specific_args(parser)
hparams = parser.parse_args()
if not torch.cuda.is_available():
hparams.gpus = None
if hparams.train:
_train(hparams)
def add_model_specific_args(
cls, parser: HyperOptArgumentParser) -> HyperOptArgumentParser:
""" Parser for Estimator specific arguments/hyperparameters.
:param parser: HyperOptArgumentParser obj
Returns:
- updated parser
"""
parser.add_argument(
"--encoder_model",
default="bert-base-uncased",
type=str,
help="Encoder model to be used.",
)
parser.add_argument(
"--encoder_learning_rate",
default=1e-05,
type=float,
help="Encoder specific learning rate.",
)
parser.add_argument(
"--learning_rate",
default=3e-05,
type=float,
help="Classification head learning rate.",
)
parser.opt_list(
"--nr_frozen_epochs",
default=1,
type=int,
help="Number of epochs we want to keep the encoder model frozen.",
tunable=True,
options=[0, 1, 2, 3, 4, 5],
)
# Data Args:
parser.add_argument(
"--label_set",
default="pos,neg",
type=str,
help="Classification labels set.",
)
parser.add_argument(
"--train_csv",
default="data/imdb_reviews_train.csv",
type=str,
help="Path to the file containing the train data.",
)
parser.add_argument(
"--dev_csv",
default="data/imdb_reviews_test.csv",
type=str,
help="Path to the file containing the dev data.",
)
parser.add_argument(
"--test_csv",
default="data/imdb_reviews_test.csv",
type=str,
help="Path to the file containing the dev data.",
)
parser.add_argument(
"--loader_workers",
default=8,
type=int,
help="How many subprocesses to use for data loading. 0 means that \
the data will be loaded in the main process.",
)
return parser
# run hopt
# creates and submits jobs to slurm
cluster.optimize_parallel_cluster_gpu(
main,
nb_trials=24,
job_name='grid_test'
)
if __name__ == '__main__':
# subclass of argparse
parser = HyperOptArgumentParser(strategy='grid_search')
# params we don't want to search over. will use the default value
parser.add_argument('--slurm_log_path', default=os.getcwd(), type=str, help='Save experiments here')
parser.add_argument('--vocab_size', default=3432, type=int)
parser.add_argument('--embed_size', default=150, type=int)
parser.add_argument('--hidden_size', default=250, type=int)
parser.add_argument('--num_workers', default=4, type=int)
parser.add_argument('--num_layers', default=2, type=int)
parser.add_argument('--batch_size', default=32, type=int)
# params we want to search over
#parser.opt_list('--embed_size', default=100, type=int, tunable=True, options=[150, 200, 250, 300])
#arser.opt_list('--hidden_size', default=100, type=int, tunable=True, options=[100, 200, 300])
#parser.opt_list('--num_layers', default=1, type=int, tunable=True, options=[1, 2])
#parser.opt_list('--batch_size', default=32, type=int, tunable=True, options=[32, 64, 128])
#parser.opt_list('--num_workers', default=15, type=int, tunable=True, options=[15])
def add_model_specific_args(parent_parser, root_dir): # pragma: no cover
parser = HyperOptArgumentParser(strategy=parent_parser.strategy,
parents=[parent_parser])
parser.set_defaults(device=torch.device(
'cuda' if torch.cuda.is_available() else 'cpu'))
# network params
parser.opt_list('--gcn_mid_dim',
default=256,
type=int,
options=[128, 256, 512, 1024],
tunable=True)
parser.opt_list('--gcn_output_dim',
default=256,
type=int,
options=[128, 256, 512, 1024],
tunable=True)
parser.opt_list('--txtcnn_drop_prob',
default=0.0,
options=[0.0, 0.1, 0.2],
type=float,
tunable=True)
parser.opt_list('--gcn_drop_prob',
default=0.5,
options=[0.2, 0.5],
type=float,
tunable=True)
parser.opt_list('--warploss_margin',
default=0.4,
type=float,
tunable=True)
parser.opt_list('--freeze_embeddings',
default=True,
options=[True, False],
type=lambda x: (str(x).lower() == 'true'),
tunable=True)
parser.opt_list('--txtcnn_pfilter_num1',
default=64,
options=[16, 32, 64, 128],
type=int,
tunable=True)
parser.opt_list('--txtcnn_pfilter_num2',
default=64,
options=[16, 32, 64, 128],
type=int,
tunable=True)
parser.opt_list('--txtcnn_pfilter_num3',
default=64,
options=[16, 32, 64, 128],
type=int,
tunable=True)
parser.opt_list('--txtcnn_pfilter_num4',
default=64,
options=[16, 32, 64, 128],
type=int,
tunable=True)
parser.opt_list('--txtcnn_rfilter_num1',
default=64,
options=[16, 32, 64, 128],
type=int,
tunable=True)
parser.opt_list('--txtcnn_rfilter_num2',
default=32,
options=[16, 32, 64, 128],
type=int,
tunable=True)
# data
parser.add_argument('--data_root',
default=os.path.join(root_dir, 'data'),
type=str)
parser.add_argument('--top_t', default=6, type=int)
parser.add_argument('--total_onehop', default=20, type=int)
parser.add_argument('--total', default=50, type=int)
parser.add_argument('--shuffle',
default=True,
type=lambda x: (str(x).lower() == 'true'))
parser.add_argument('--train_div', default=1.0, type=float)
# training params (opt)
parser.opt_list('--batch_size',
default=64,
options=[32, 64, 128, 256],
type=int,
tunable=False)
parser.opt_list('--max_nb_epochs',
default=8,
options=[256, 512, 1024],
type=int,
tunable=False)
parser.opt_list('--learning_rate',
default=0.0005,
options=[0.0001, 0.0005, 0.001],
type=float,
tunable=True)
parser.opt_list('--weight_decay',
default=0.001,
options=[0.0001, 0.0005, 0.001],
type=float,
tunable=True)
parser.add_argument('--model_save_path',
default=os.path.join(root_dir, 'experiment'),
type=str)
return parser
# Pretend to train.
x = torch.rand((1, hparams.x_val))
for train_step in range(0, 100):
y = torch.rand((hparams.x_val, 1))
out = x.mm(y)
exp.log({'fake_err': out.item()})
# Save exp when .
exp.save()
if __name__ == '__main__':
# Set up our argparser and make the y_val tunable.
parser = HyperOptArgumentParser(strategy='random_search')
parser.add_argument('--test_tube_exp_name', default='my_test')
parser.add_argument('--log_path', default='/some/path/to/log')
parser.opt_list('--y_val',
default=12,
options=[1, 2, 3, 4, 5, 6],
tunable=True)
parser.opt_list('--x_val',
default=12,
options=[20, 12, 30, 45],
tunable=True)
hyperparams = parser.parse_args()
# Enable cluster training.
cluster = SlurmCluster(hyperparam_optimizer=hyperparams,
log_path=hyperparams.log_path,
python_cmd='python3',
def add_model_specific_args(parent_parser):
parser = HyperOptArgumentParser(strategy=parent_parser.strategy,
parents=[parent_parser],
add_help=False)
parser.opt_list('--q_drop',
default=0.35,
options=[0.25, 0.3, 0.35, 0.15, 0.4, 0.1],
type=float,
tunable=False)
parser.add_argument('--q_hid_dim', default=1024, type=int)
parser.add_argument('--padding_idx', type=int, default=0)
parser.add_argument('--net_init', default='k_u')
parser.opt_list('--iter_num',
default=3,
options=[1, 2, 3, 4, 5],
type=int)
parser.add_argument('--g_init_method', default='full', type=str)
parser.add_argument('--stem_in_dim', default=2048, type=int)
parser.add_argument('--stem_out_dim', default=1024, type=int)
parser.opt_list('--stem_norm',
default='custom',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--stem_orders',
default='dlna',
options=['dlna', 'dnla'],
type=str,
tunable=False)
parser.opt_list('--stem_drop',
default=0.35,
options=[0.3, 0.35, 0.4],
type=int)
parser.opt_list('--stem_method',
default='linear',
options=['linear', 'film', 'double_linear'],
type=str)
parser.add_argument('--stem_use_act', action='store_true')
parser.opt_list('--e_f_norm',
default='weight',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--e_f_orders',
default='dlna',
options=['dlna', 'dnla'],
type=str,
tunable=False)
parser.opt_list('--e_f_drop',
default=0.35,
options=[0.3, 0.35, 0.4],
type=int)
parser.add_argument('--e_dim', default=512, type=int)
parser.opt_list('--e_f_method',
default='mul_film',
options=['mul_film', 'cat_film'],
type=str)
parser.add_argument('--e_f_use_nGeo',
action='store_true',
default=True)
parser.opt_list('--e_w_method',
default='linear_softmax_8',
options=['linear_softmax_10', 'linear_softmax_12'],
type=str)
parser.opt_list('--e_w_norm',
default='weight',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--e_w_orders',
default='dlna',
options=['dlna', 'dnla'],
type=str,
tunable=False)
parser.opt_list('--e_w_drop',
default=0.35,
options=[0.3, 0.35, 0.4],
type=int)
parser.opt_list('--e_p_method',
default='linear',
options=['linear', 'share'],
type=str)
parser.opt_list('--e_p_norm',
default='weight',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--e_p_orders',
default='dlna',
options=['dlna', 'dnla'],
type=str,
tunable=False)
parser.opt_list('--e_p_drop',
default=0.35,
options=[0.3, 0.35, 0.4],
type=int)
parser.opt_list('--e_p_act',
default='relu',
options=['relu', 'swish', 'elu'],
type=str,
tunable=False)
parser.opt_list('--n_f_method',
default='film',
options=['linear', 'film'],
type=str)
parser.opt_list('--n_f_drop',
default=0.35,
options=[0.3, 0.35, 0.4],
type=int)
parser.opt_list('--n_geo_method',
default='cat',
options=['cat', 'sum', 'linear_cat'],
type=str)
parser.add_argument('--n_geo_reuse', action='store_true', default=True)
parser.add_argument('--n_geo_dim', default=64, type=int)
parser.add_argument('--n_geo_out_dim', default=64, type=int)
parser.opt_list('--n_geo_norm',
default='weight',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--n_geo_orders',
default='lna',
options=['lna', 'nla'],
type=str,
tunable=False)
parser.opt_list('--e_geo_method',
default='linear',
options=['linear', 'cat'],
type=str)
parser.add_argument('--e_geo_dim', default=128, type=int)
parser.add_argument('--e_geo_out_dim', default=128, type=int)
parser.add_argument('--e_geo_reuse', action='store_true', default=True)
parser.add_argument('--e_geo_aug', action='store_true')
parser.opt_list('--e_geo_norm',
default='weight',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--e_geo_orders',
default='lna',
options=['lna', 'nla'],
type=str,
tunable=False)
parser.opt_list('--cls_method',
default='linear',
options=['linear'],
type=str)
parser.opt_list('--cls_norm',
default='weight',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--cls_orders',
default='dlna',
options=['dlna', 'dnla'],
type=str,
tunable=False)
parser.opt_list('--cls_drop',
default=0.35,
options=[0.3, 0.35, 0.4],
type=int)
parser.opt_list('--cls_act',
default='relu',
options=['relu', 'swish'],
type=str,
tunable=False)
parser.opt_list('--f_c_norm',
type=str,
default='weight',
options=['weight', 'custom', 'layer'])
parser.opt_list('--f_c_drop',
default=0.,
options=[0.3, 0.35, 0.4],
type=int)
parser.opt_list('--f_c_orders',
default='dln',
options=['dln', 'dnl'],
type=str,
tunable=False)
parser.opt_list('--f_x_norm',
default='layer',
options=['custom', 'layer', 'weight'],
type=str,
tunable=False)
parser.opt_list('--f_x_orders',
default='dln',
options=['dln', 'dnl'],
type=str,
tunable=False)
parser.add_argument('--f_x_norm_affine', action='store_true')
parser.opt_list('--f_act',
default='relu',
options=['relu', 'swish', 'elu'],
type=str,
tunable=False)
parser.opt_list('--pool_method',
default='mix',
options=['mix', 'mean', 'max'],
type=str)
parser.opt_list('--lr',
default=2.5e-4,
type=float,
options=[2.5e-4, 3.5e-4],
tunable=False)
parser.opt_list('--sched_factor',
default=0.5,
type=float,
options=[0.1, 0.8, 0.6, 0.3],
tunable=False)
parser.opt_list('--optimizer',
type=str,
default='adam',
options=['adam', 'sgd'])
parser.opt_list('--sched',
type=str,
default='plat',
options=['plat', 'cyclic', 'sgdr'])
parser.opt_list('--dataset',
default='vqa2_cp',
type=str,
options=['vaq2_cp', 'gqa_lcgn'])
parser.add_argument('--epochs', default=12, type=int)
parser.add_argument('--batch_size', default=64, type=int)
parser.add_argument('--num_workers', default=5, type=int)
parser.add_argument('--grad_clip', default=0., type=float)
parser.add_argument('--weight_decay', default=0., type=float)
parser.add_argument('--use_warmup', action='store_true')
return parser
if __name__ == '__main__':
# dirs
root_dir = os.path.dirname(os.path.realpath(__file__))
demo_log_dir = os.path.join(root_dir, 'pt_lightning_demo_logs')
checkpoint_dir = os.path.join(demo_log_dir, 'model_weights')
test_tube_dir = os.path.join(demo_log_dir, 'test_tube_data')
# although we user hyperOptParser, we are using it only as argparse right now
parent_parser = HyperOptArgumentParser(strategy='grid_search',
add_help=False)
# gpu args
parent_parser.add_argument('--test_tube_save_path',
type=str,
default=test_tube_dir,
help='where to save logs')
parent_parser.add_argument('--model_save_path',
type=str,
default=checkpoint_dir,
help='where to save model')
parent_parser.add_argument('--experiment_name',
type=str,
default='pt_lightning_exp_a',
help='test tube exp name')
# allow model to overwrite or extend args
parser = LightningTemplateModel.add_model_specific_args(
parent_parser, root_dir)
hyperparams = parser.parse_args()
options=[20, 30, 40, 50],
type=int,
tunnable=True)
parser.add_opt_argument_list('--embedding_dim',
default=320,
options=[100, 200, 320, 400],
type=int,
tunnable=True)
parser.add_opt_argument_list('--max_seq_len',
default=50,
options=[50, 70, 90, 110],
type=int,
tunnable=True)
# training params
parser.add_argument('--nb_epochs', default=1000, type=int)
parser.add_argument('--optimizer_name', default='adam')
parser.add_argument('--eval_every_n_batches', default=200, type=int)
parser.add_argument('--save_every_n_batches', default=10000, type=int)
parser.add_argument('--train_mode', default='train')
# model params
parser.add_argument('--nb_grams', default=2, type=int)
# path params
parser.add_argument('--root_dir', default='./data')
parser.add_argument('--dataset_train_path', default='./data/train.csv')
parser.add_argument('--dataset_test_path', default='./data/test.csv')
parser.add_argument('--dataset_val_path', default='./data/val.csv')
parser.add_argument('--vocab_path', default='./data/vocabulary.txt')
def add_model_specific_args(parent_parser):
parser = HyperOptArgumentParser(parents=[parent_parser])
parser.add_argument('--num_freq_mask', default=1, type=int)
parser.add_argument('--num_time_mask', default=1, type=int)
parser.add_argument('--freq_mask_length', default=10, type=int)
parser.add_argument('--time_mask_length', default=20, type=int)
parser.add_argument('--feature_dim', default=320, type=int)
parser.add_argument('--model_size', default=512, type=int)
parser.add_argument('--feed_forward_size', default=512, type=int)
parser.add_argument('--hidden_size', default=64, type=int)
parser.add_argument('--dropout', default=0.1, type=float)
parser.add_argument('--num_head', default=4, type=int)
parser.add_argument('--num_layer', default=4, type=int)
parser.add_argument('--vocab_path',
default='testing_vocab.model',
type=str)
parser.add_argument('--max_feature_length', default=1024, type=int)
parser.add_argument('--max_token_length', default=50, type=int)
parser.add_argument('--lr', default=3e-4, type=float)
parser.add_argument('--enable_spec_augment', default=False, type=bool)
parser.add_argument('--train_batch_size', default=16, type=int)
parser.add_argument('--train_loader_num_workers', default=4, type=int)
parser.add_argument('--val_batch_size', default=16, type=int)
parser.add_argument('--val_loader_num_workers', default=4, type=int)
return parser
GPUs = ['0', '1', '2', '4']
os.environ["CUDA_VISIBLE_DEVICES"] = GPUs[job_index_nb]
main_trainer(trial_params)
if __name__ == '__main__':
parser = HyperOptArgumentParser(strategy='random_search')
parser.add_opt_argument_list('--time_steps', default=3, options=[3, 5, 7, 9, 11], type=int, help='number of sequential readings', tunnable=False)
# model params
parser.add_opt_argument_list('--gamma', default=0.00001, options=[0.00001], type=int, tunnable=False)
parser.add_opt_argument_list('--kernel', default='rbf', options=['rbf'], type=str, tunnable=False)
parser.add_opt_argument_list('--C', default=4, options=[4, 5, 6], type=int, tunnable=False)
parser.add_argument('--nb_trials', default=200, type=int)
# path vars
parser.add_argument('--data_path', default='/Users/waf/Developer/temp_floor/floor/data/in_out_classifier_data/data')
parser.add_argument('--model_save_path', default='/Users/waf/Developer/temp_floor/floor/logs/weights/svm')
parser.add_argument('--tt_save_path', default='/Users/waf/Developer/temp_floor/floor/logs/training_logs/svm')
parser.add_argument('--tt_name', default='svm_final_1')
parser.add_argument('--tt_description', default='hyperopt')
parser.add_argument('--debug', default=False, type=bool)
parser.add_argument('--local', default=True, type=bool)
parser.add_json_config_argument('--config', default='/Users/waf/Developer/temp_floor/floor/logs/run_configs/local.json')
hyperparams = parser.parse_args()
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
if hyperparams.local:
feed_dict = {input_x: text_enc, input_y: labels_enc}
out_value, out_score = sess.run([value, score], feed_dict=feed_dict)
print('label: ', labels[out_value])
print('score: ', out_score)
if __name__ == '__main__':
parser = HyperOptArgumentParser(strategy='random_search')
# path params
parser.add_opt_argument_list('--lr_1', default=0.0001, options=[0.0001, 0.0002, 0.0004, 0.0008, 0.001, 0.002],
type=float,
tunnable=True)
parser.add_opt_argument_list('--batch_size', default=10, options=[20, 30, 40, 50], type=int, tunnable=True)
parser.add_opt_argument_list('--embedding_dim', default=320, options=[100, 200, 320, 400], type=int, tunnable=True)
parser.add_argument('--root_dir', default='./data')
parser.add_argument('--dataset_train_path', default='./data/train.csv')
parser.add_argument('--dataset_test_path', default='./data/test.csv')
parser.add_argument('--dataset_val_path', default='./data/val.csv')
parser.add_argument('--vocab_path', default='./data/vocabulary.txt')
parser.add_opt_argument_list('--max_seq_len', default=50, options=[50, 70, 90, 110], type=int, tunnable=True)
parser.add_argument('--epoch', default=0)
parser.add_argument('--step', default='FINAL')
parser.add_argument('--model_save_dir', default='./experiments')
parser.add_argument('--test_tube_dir', default='')
# experiment params
parser.add_argument('--exp_name', default='dual_conv_dense')
hparams = parser.parse_args()
test_main(hparams=hparams)
checkpoint_path = experiment_folder + "/checkpoints/" + checkpoints[-1]
model = BERTClassifier.load_from_metrics(weights_path=checkpoint_path,
tags_csv=tags_csv_file)
# Make sure model is in prediction mode
model.eval()
model.freeze()
return model
if __name__ == "__main__":
parser = HyperOptArgumentParser(description="Minimalist BERT Classifier",
add_help=True)
parser.add_argument(
"--experiment",
default=
'../../data/experiments/lightning_logs/version_29-04-2020--22-14-10',
type=str,
help="Path to the experiment folder.")
hparams = parser.parse_args()
print("Loading model...")
model = load_model_from_experiment(hparams.experiment)
print(model)
print("Please write a movie review or quit to exit the interactive shell:")
# Get input sentence
test_path = '../../data/output/test.csv'
test = pd.read_csv(test_path, encoding='utf-8')
labels = []
for index in range(test.shape[0]):
