def _test_fit_loader(device):
seed_all(42)
save_dir = tmpdir.mkdir("/test_save")
file_name = "test_classification.bin"
dataset = torch.utils.data.TensorDataset(X, y)
test_dataset = create_ds((test_data, ))
loader = torch.utils.data.DataLoader(dataset, batch_size=32)
test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=16)
metric_list = [
Accuracy(num_classes=num_classes),
]
callbacks = [
cbs.EarlyStopping(monitor="val_accuracy", mode="max"),
cbs.ModelCheckpoint(monitor="val_accuracy",
mode="max",
save_dir=save_dir,
file_name=file_name),
cbs.CosineAnnealingWarmRestarts(T_0=2),
]
exp = Experiment(
num_epochs=10,
fp16=fp16,
device=device,
seed=42,
)
exp.compile_experiment(
model_config=config,
metrics=metric_list,
callbacks=callbacks,
main_metric="accuracy",
)
assert isinstance(exp.state, State) is True
exp.fit_loader(train_dl=loader, valid_dl=loader)
logs = exp.get_logs()
assert isinstance(logs, pd.DataFrame) is True
outputs = []
for op in exp.predict_on_loader(test_dl=test_dl,
device=device,
path_to_model=os.path.join(
save_dir, file_name)):
outputs.extend(op)
assert len(outputs) == test_samples
def _test_fit_loader(device):
seed_all(42)
save_dir = tmpdir.mkdir("/test_save")
file_name = "test_classification.bin"
dataset = torch.utils.data.TensorDataset(X, y)
test_dataset = torch.utils.data.TensorDataset(test_data)
loader = torch.utils.data.DataLoader(dataset, batch_size=32)
test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=16)
model = torch.nn.Linear(num_features, num_classes)
metric_list = [
metrics.Accuracy(num_classes=num_classes, multilabel=False),
metrics.F1Score(num_classes=num_classes, multilabel=False),
]
callbacks = [
cbs.EarlyStopping(monitor="accuracy", mode="max"),
cbs.ModelCheckpoint(monitor="accuracy", mode="max", save_dir=save_dir, file_name=file_name),
cbs.CosineAnnealingWarmRestarts(T_0=2),
]
exp = Experiment(
num_epochs=10,
fp16=fp16,
device=device,
seed=42,
)
exp.compile_experiment(
model=model,
metrics=metric_list,
callbacks=callbacks,
main_metric="accuracy",
optimizer=optimizer,
optimizer_params=optimizer_params,
criterion=criterion,
)
exp.fit_loader(train_dl=loader, valid_dl=loader)
exp.plot_history(keys=["accuracy"], save_fig=False, plot_fig=False)
outputs = []
for op in exp.predict_on_loader(
test_dl=test_dl, device=device, path_to_model=os.path.join(save_dir, file_name)
):
outputs.extend(op)
assert len(outputs) == test_samples
"mlp_params": {"projection_size": 256, "hidden_size": 1024},
},
"predictor": {
"in_channels": 256,
"projection_size": 256,
"hidden_size": 1024,
},
},
optimizer="AdamW",
optimizer_params={"lr": 1e-4, "weight_decay": 1e-3},
criterion=normalized_mse,
)
# Defining callbacks.
callbacks = [
cbs.ModelCheckpoint(save_dir="./", mode="min", monitor="train_loss"),
init_target_network,
cbs.CosineAnnealingWarmRestarts(T_0=2),
]
# Compiling and running the experiment.
byol_exp = BYOLExperiment(
num_epochs=25,
seed=42,
device="cuda",
fp16=True,
momentum=4e-3,
augmentation_fn=default_augmentation,
image_size=(96, 96),
)
)
valid_dl = (
TextDataset.from_df(df=valid_df, tokenizer=tokenizer, max_len=128, input_columns=["text"])
.targets_from_df(target_columns=["label"])
.batch(batch_size=16, shuffle=True)
)
# Defining metrics
metric_list = [torchmetrics.Accuracy(threshold=0.6)]
# Defining Callbacks.
callbacks = [
cbs.EarlyStopping(monitor="val_accuracy", patience=2, mode="max"),
cbs.ModelCheckpoint(
monitor="val_accuracy", mode="max", save_dir="./", file_name="model.bin"
),
cbs.ReduceLROnPlateau(mode="max", patience=2),
]
config = ModelConfig(
nn_module=Model,
module_params={"dropout": 0.3, "out_features": 1},
optimizer="AdamW",
optimizer_params={"lr": 3e-4},
criterion=crit.BCEWithLogitsFlat,
)
# Compiling and Running Experiment.
exp = Experiment(
num_epochs=3,
random_state=123,
test_size=0.20)
train_images = train_images.reshape(train_images.shape[0], 1, 28, 28)
val_images = val_images.reshape(val_images.shape[0], 1, 28, 28)
train_images = train_images / 255.0
val_images = val_images / 255.0
# Defining Metrics
metric_list = [Accuracy(num_classes=classes, average="micro")]
# Defining Callbacks
callbacks = [
cbs.EarlyStopping(monitor="val_accuracy", mode="max", patience=5),
cbs.ModelCheckpoint(monitor="val_accuracy", mode="max"),
cbs.ReduceLROnPlateau(mode="max", patience=2),
cbs.DiscordNotifierCallback(
exp_name="MNIST-EXP",
webhook_url=os.environ.get("DISCORD_WEBHOOK")),
]
# Defining ModelConfig for Experiment
config = ModelConfig(
nn_module=Net, # The uninstantiated model_class for the neural network.
module_params={"out_features": 10},
optimizer="Adam",
optimizer_params={"lr": 3e-4},
criterion="cross_entropy",
)
if __name__ == "__main__":
# Defining ModelConfig for experiment
exp_config = ModelConfig(
nn_module={"discriminator": Discriminator, "generator": Generator},
module_params={
"discriminator": {"output_dim": 1},
"generator": {"latent_dim": 16},
},
optimizer={"discriminator": "Adam", "generator": "Adam"},
optimizer_params={"discriminator": dict(lr=1e-3), "generator": dict(lr=2e-4)},
criterion="binary_cross_entropy",
)
# Some callbacks
callbacks = [cbs.ModelCheckpoint(mode="min", monitor="train_loss_g", save_dir="./")]
transform = tv.transforms.Compose(
[
tv.transforms.Grayscale(num_output_channels=1),
tv.transforms.ToTensor(),
tv.transforms.Normalize((0.5,), (0.5,)),
]
)
# Creating dataloaders
dataset = ImageFolder(root=os.path.join("mnist_png", "training"), transform=transform)
dataloader = torch.utils.data.DataLoader(dataset, batch_size=16)
# Compiling and Running the experiment.
exp = DCGANExperiment(
latent_dim=16,