Steve Niafas sniafas

MLE | @mlflow Ambassador |GNU/Linux advocate

sniafas / predict.py

Created April 17, 2023 14:53

Prediction

	def predict(self, x):
	"""Define prediction"""
	print("Predicting...")
	# get testing set
	test_dataloader = self.get_data_to_inference(x, len(x))
	with torch.no_grad():
	for x in test_dataloader:
	preds = self.model(x)
	return preds

sniafas / history.py

Created April 17, 2023 14:41

Training History

	# calculate acc, loss
	epoch_loss = total_loss/batch_idx
	epoch_accuracy = total_accuracy/len(train_dataloader.dataset)

	# record training history
	history["accuracy"].append(epoch_accuracy)
	history["loss"].append(epoch_loss)

sniafas / training.py

Created April 17, 2023 14:27

Training Runtime

	def get_loss_function(self):
	"""Define Loss function"""
	return nn.CrossEntropyLoss()

	def get_optimizer(self, params):
	"""Define adam as the optimizer to minimise the loss"""
	return torch.optim.Adam(params, lr=self.lr)

	def get_data(self, x, y, batch_size):
	"""Get a data loader"""

sniafas / fc_model.py

Last active April 17, 2023 14:11

fully connected

	class fcModel(nn.Module):
	def __init__(self, input_shape, output_shape):
	"""Model class constructor"""
	super(fcModel, self).__init__()

	self.linear = nn.Linear(input_shape, 8)
	self.fully_connected_stack = nn.Sequential(
	nn.Linear(8, 8),
	nn.ReLU(),
	nn.Linear(8, 8),

sniafas / data_loaders.py

Created April 17, 2023 13:36

	class TrainingDatasetLoader(Dataset):
	def __init__(self, x, y):
	self.points = torch.Tensor(x)
	self.labels = torch.Tensor(y).type(torch.long)
	self.len = len(self.labels)

	def __getitem__(self, idx):

	point = self.points[idx]
	label = self.labels[idx]

sniafas / load_spiral.py

Created April 17, 2023 13:22

X, Y = datasets.data_spiral(10000, 50)

sniafas / fit.py

Last active April 5, 2023 08:57

logistic fit

	def fit(self, data):
	"""Train and evaluate model"""

	optimizer = self.get_optimizer(self.model.parameters())
	train_dataloader = self.get_data(data, self.batch_size, True)
	print("Training...")
	for epoch in range(1, self.epochs+1):

	for batch_idx, (x, y) in enumerate(train_dataloader):

sniafas / data_loader.py

Last active April 5, 2023 09:05

Data Loader

	from torch.utils.data import Dataset, DataLoader

	class ImageDataLoader(Dataset):
	def __init__(self, data):
	self.images = data[0]
	self.labels = data[1]
	self.len = len(self.labels)

	def __getitem__(self, idx):
	image = self.images[idx]

sniafas / training_loop.py

Last active April 4, 2023 22:11

Pytorch - Linear Regression - Training Loop

	def get_weight_biases():
	return w[0].item(), b[0].item()

	# initialise model
	model = Model(x.shape[1], y.shape[1])

	# initialise optimizer
	optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

	# initialise weights & biases

sniafas / loss_optimizer.py

Created March 31, 2023 15:48

Pytorch - Loss Function & Optimizer

	loss_function = nn.MSELoss()
	optimizer = torch.optim.SGD(model.parameters(), lr=0.01)