Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

timing log and separate optim for g and d #35

Open
wants to merge 2 commits into
base: development
Choose a base branch
from
Open

timing log and separate optim for g and d #35

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d25ea25
timing log and separate optim for g and d
AnniLi1212 Feb 14, 2024
89e63c8
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Feb 14, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
81 changes: 66 additions & 15 deletions setup_training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,7 +204,26 @@ def parse_optimization_args(parser):
type=str,
default="rmsprop",
help="pick optimizer",
choices=["adam", "rmsprop", "adadelta", "agcd"],
choices=["adam", "rmsprop", "adadelta", "agcd", "sgd"],
)
parser.add_argument(
"--use_different_optimizers",
action="store_true",
help="Use different optimizers for generator and discriminator",
)
parser.add_argument(
"--optimizer-G",
type=str,
default="rmsprop",
help="pick optimizer for generator",
choices=["adam", "rmsprop", "adadelta", "agcd", "sgd"],
)
parser.add_argument(
"--optimizer-D",
type=str,
default="rmsprop",
help="pick optimizer for discriminator",
choices=["adam", "rmsprop", "adadelta", "agcd", "sgd"],
)
parser.add_argument(
"--loss",
Expand Down Expand Up @@ -248,6 +267,12 @@ def parse_optimization_args(parser):
default=1,
help="number of generator updates for each critic update (num-critic must be 1 for this to apply)",
)
parser.add_argument(
"--sgd-momentum",
type=float,
default=0.9,
help="momentum for the SGD optimizer",
)


def parse_regularization_args(parser):
Expand Down Expand Up @@ -1410,7 +1435,7 @@ def models(args, gen_only=False):
from mpgan import Graph_GAN

G = Graph_GAN(gen=False, args=deepcopy(args))

logging.info(f"# of parameters in D: {count_parameters(D)}")

if args.load_model:
Expand Down Expand Up @@ -1518,19 +1543,45 @@ def optimizers(args, G, D):
else:
D_params = D.parameters()

if args.optimizer == "rmsprop":
G_optimizer = optim.RMSprop(G_params, lr=args.lr_gen)
D_optimizer = optim.RMSprop(D_params, lr=args.lr_disc)
elif args.optimizer == "adadelta":
G_optimizer = optim.Adadelta(G_params, lr=args.lr_gen)
D_optimizer = optim.Adadelta(D_params, lr=args.lr_disc)
elif args.optimizer == "adam" or args.optimizer == "None":
G_optimizer = optim.Adam(
G_params, lr=args.lr_gen, weight_decay=5e-4, betas=(args.beta1, args.beta2)
)
D_optimizer = optim.Adam(
D_params, lr=args.lr_disc, weight_decay=5e-4, betas=(args.beta1, args.beta2)
)
if args.use_different_optimizers:
if args.optimizer_G == "rmsprop":
G_optimizer = optim.RMSprop(G_params, lr=args.lr_gen)
elif args.optimizer_G == "adadelta":
G_optimizer = optim.Adadelta(G_params, lr=args.lr_gen)
elif args.optimizer_G == "adam":
G_optimizer = optim.Adam(
G_params, lr=args.lr_gen, weight_decay=5e-4, betas=(args.beta1, args.beta2)
)
elif args.optimizer_G == "sgd":
G_optimizer = optim.SGD(G_params, lr=args.lr_gen, momentum=args.sgd_momentum)

if args.optimizer_D == "rmsprop":
D_optimizer = optim.RMSprop(D_params, lr=args.lr_disc)
elif args.optimizer_D == "adadelta":
D_optimizer = optim.Adadelta(D_params, lr=args.lr_disc)
elif args.optimizer_D == "adam":
D_optimizer = optim.Adam(
D_params, lr=args.lr_disc, weight_decay=5e-4, betas=(args.beta1, args.beta2)
)
elif args.optimizer_D == "sgd":
D_optimizer = optim.SGD(D_params, lr=args.lr_disc, momentum=args.sgd_momentum)
else:
if args.optimizer == "rmsprop":
G_optimizer = optim.RMSprop(G_params, lr=args.lr_gen)
D_optimizer = optim.RMSprop(D_params, lr=args.lr_disc)
elif args.optimizer == "adadelta":
G_optimizer = optim.Adadelta(G_params, lr=args.lr_gen)
D_optimizer = optim.Adadelta(D_params, lr=args.lr_disc)
elif args.optimizer == "adam" or args.optimizer == "None":
G_optimizer = optim.Adam(
G_params, lr=args.lr_gen, weight_decay=5e-4, betas=(args.beta1, args.beta2)
)
D_optimizer = optim.Adam(
D_params, lr=args.lr_disc, weight_decay=5e-4, betas=(args.beta1, args.beta2)
)
elif args.optimizer == "sgd":
G_optimizer = optim.SGD(G_params, lr=args.lr_gen, momentum=args.sgd_momentum)
D_optimizer = optim.SGD(D_params, lr=args.lr_disc, momentum=args.sgd_momentum)

if args.load_model:
G_optimizer.load_state_dict(
Expand Down
Loading