def main():
structures = [(8, 3), (1, 27)] #(cardinality, layers)
substructures = ["ResNeXt3", "RDB3"]
blocks = ["NeXt", "vanilla"]
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
idx = 0
for block in blocks:
for substruct in substructures:
for struct in structures:
idx += 1
#split work across 2 gpus:
if idx - 1 < 6 and GPU_DEVICE == 1:
continue
elif idx - 1 >= 4 and GPU_DEVICE == 0:
continue
(cardinality, layers) = struct
kwargs = {
"layers": layers,
"cardinality": cardinality,
"block_type": block,
"module_type": substruct
}
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
kwargs = {
"cardinality": 1,
"block_type": "RNAB",
"sigmoid": True,
"module_type": "Bespoke",
"attenuation": False
}
layers = [
1,
2,
4,
8,
]
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
idx = 0
for layer in layers:
kwargs[
"subBlock"] = "NeXt" #this performed slightly better on first case
kwargs["layers"] = layer
idx += 1
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
layers = [3, 6, 9, 18, 27]
cardinalities = [1, 4, 8, 16, 32]
substructure = "ResNeXt3"
block = "NeXt"
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
idx = 0
for layer in layers:
for cardinality in cardinalities:
idx_ = idx
idx += 1
#split work across 4 gpus:
if idx_ % 4 != GPU_DEVICE:
continue
kwargs = {
"layers": layer,
"cardinality": cardinality,
"block_type": block,
"module_type": substructure
}
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx_) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
activations = ["relu"]
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
expt.LR *= 0.1
idx = 0
for act in activations:
kwargs = {
"layers": 0,
"cardinality": 2,
"aug_scheme": 4,
"activation": act
}
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
blocks = ["CBAM_vanilla", "vanilla", "CBAM_NeXt"]
kwargs = {"cardinality": 1, "block_type": "RAB",
"sigmoid": True, "module_type": "Bespoke",
"attenuation": False, "layers": 4,
"aug_scheme": 0}
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
idx = 0
for block in blocks:
kwargs["subBlock"] = block
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx_) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS, test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate = expt.LR, print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
blocks = ["NeXt", "vanilla"]
kwargs = {
"layers": 1,
"cardinality": 1,
"block_type": "RNAB",
"module_type": "Bespoke"
}
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
idx = 0
for block in blocks:
kwargs["subBlock"] = block
idx += 1
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
exp_base = "experiments/train2/07c/"
kwargs = {
"model_name": "Tucodec",
"block_type": "NeXt",
"Cstd": 64,
"sigmoid": False,
"activation": "prelu"
}
aug_schemes = list(range(5, 7))
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
expt.EPOCHS = 300
idx = 0
for aug_scheme in aug_schemes:
kwargs["aug_scheme"] = aug_scheme
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = CLIC(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
exp_base = "experiments/train2/09b/"
activations = ["relu", "GDN"]
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
expt.EPOCHS = 150
for act in activations:
kwargs = {
"layers": 0,
"cardinality": 2,
"aug_scheme": 0,
"activation": act
}
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
def main():
idx = 0
##################### 02c
structure = (4, 27) #(cardinality, layers)
substructures = ["ResNeXt3", "RDB3"]
blocks = ["CBAM_vanilla", "vanilla", "NeXt", "CBAM_NeXt"]
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
expt.EPOCHS = 150
for substruct in substructures:
for block in blocks:
(cardinality, layers) = structure
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
kwargs = {
"layers": layers,
"cardinality": cardinality,
"block_type": block,
"module_type": substruct,
"aug_scheme": 0
}
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx_) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
################# 03c
blocks = ["CBAM_vanilla", "vanilla", "CBAM_NeXt"]
kwargs = {
"cardinality": 1,
"block_type": "RAB",
"sigmoid": True,
"module_type": "Bespoke",
"attenuation": False,
"layers": 4,
"aug_scheme": 0
}
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
expt.EPOCHS = 150
for block in blocks:
kwargs["subBlock"] = block
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = ResStack3(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx_) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)
################################06a5
blocks = [
"NeXt",
"vanilla",
"CBAM_vanilla",
"CBAM_NeXt",
]
Cstd = 64
sigmoid = False
activations = ["relu"]
if TEST:
expt = ExptConfigTest()
else:
expt = ExptConfig()
expt.EPOCHS = 300
for activation in activations:
for block in blocks:
kwargs = {
"model_name": "Tucodec",
"block_type": block,
"Cstd": Cstd,
"sigmoid": sigmoid,
"activation": activation,
"aug_scheme": 0
}
idx_ = idx
idx += 1
if idx_ % NUM_GPU != GPU_DEVICE:
continue
for k, v in kwargs.items():
print("{}={}, ".format(k, v), end="")
print()
settings = CLIC(**kwargs)
settings.GPU_DEVICE = GPU_DEVICE
settings.export_env_vars()
expdir = exp_base + str(idx - 1) + "/"
trainer = TrainAE(settings, expdir, expt.calc_DA_MAE)
expdir = trainer.expdir #get full path
model = trainer.train(expt.EPOCHS,
test_every=expt.test_every,
num_epochs_cv=expt.num_epochs_cv,
learning_rate=expt.LR,
print_every=expt.print_every,
small_debug=expt.SMALL_DEBUG_DOM)