Overview:
- Tensor Basics
- Create, Operations, NumPy, GPU Support
- Autograd
- Linear regression example
- Training Loop with: Model, Loss & Optimizer
- A typical PyTorch training pipeline
- Neural Network
- Also: GPU, Datasets, DataLoader, Transforms & Evaluation
- Convolutional Neural Network
- Also: Save/Load model
1. Tensors
Everything in PyTorch is based on Tensor operations. A Tensor is a multi-dimensional matrix containing elements of a single data type:
import torch
# torch.empty(size): uninitiallized
x = torch.empty(1) # scalar
print("empty(1):", x)
x = torch.empty(3) # vector
print("empty(3):",x)
x = torch.empty(2, 3) # matrix
print("empty(2,3):",x)
x = torch.empty(2, 2, 3) # tensor, 3 dimensions
#x = torch.empty(2,2,2,3) # tensor, 4 dimensions
print("empty(2, 2, 3):",x)
# torch.rand(size): random numbers [0, 1]
x = torch.rand(5, 3)
print("rand(5,3):", x)
# torch.zeros(size), fill with 0
# torch.ones(size), fill with 1
x = torch.zeros(5, 3)
print("zeros(5,3):", x)
# check size
print("size", x.size()) # x.size(0)
print("shape", x.shape) # x.shape[0]
# check data type
print(x.dtype)
# specify types, float32 default
x = torch.zeros(5, 3, dtype=torch.float16)
print(x)
# check type
print(x.dtype)
# construct from data
x = torch.tensor([5.5, 3])
print(x, x.dtype)
# requires_grad argument
# This will tell pytorch that it will need to calculate the gradients for this tensor
# later in your optimization steps
# i.e. this is a variable in your model that you want to optimize
x = torch.tensor([5.5, 3], requires_grad=True)
print(x)
Operations with Tensors
# Operations
x = torch.ones(2, 2)
y = torch.rand(2, 2)
# elementwise addition
z = x + y
# torch.add(x,y)
# in place addition, everythin with a trailing underscore is an inplace operation
# i.e. it will modify the variable
# y.add_(x)
print(x)
print(y)
print(z)
# subtraction
z = x - y
z = torch.sub(x, y)
# multiplication
z = x * y
z = torch.mul(x,y)
# division
z = x / y
z = torch.div(x,y)
# Slicing
x = torch.rand(5,3)
print(x)
print("x[:, 0]", x[:, 0]) # all rows, column 0
print("x[1, :]", x[1, :]) # row 1, all columns
print("x[1, 1]", x[1,1]) # element at 1, 1
# Get the actual value if only 1 element in your tensor
print("x[1,1].item()", x[1,1].item())
# Reshape with torch.view()
x = torch.randn(4, 4)
y = x.view(16)
z = x.view(-1, 8) # the size -1 is inferred from other dimensions
# if -1 it pytorch will automatically determine the necessary size
print(x.size(), y.size(), z.size())
NumPy
Converting a Torch Tensor to a NumPy array and vice versa is very easy
a = torch.ones(5)
print(a)
# torch to numpy with .numpy()
b = a.numpy()
print(b)
print(type(b))
# Careful: If the Tensor is on the CPU (not the GPU),
# both objects will share the same memory location, so changing one
# will also change the other
a.add_(1)
print(a)
print(b)
# numpy to torch with .from_numpy(x), or torch.tensor() to copy it
import numpy as np
a = np.ones(5)
b = torch.from_numpy(a)
c = torch.tensor(a)
print(a)
print(b)
print(c)
# again be careful when modifying
a += 1
print(a)
print(b)
print(c)
GPU Support
By default all tensors are created on the CPU. But we can also move them to the GPU (if it's available ), or create them directly on the GPU.
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
x = torch.rand(2,2).to(device) # move tensors to GPU device
#x = x.to("cpu")
#x = x.to("cuda")
x = torch.rand(2,2, device=device) # or directy create them on GPU
2. Autograd
The autograd package provides automatic differentiation for all operations on Tensors. Generally speaking, torch.autograd is an engine for computing the vector-Jacobian product. It computes partial derivates while applying the chain rule.
Set requires_grad = True
:
import torch
# requires_grad = True -> tracks all operations on the tensor.
x = torch.randn(3, requires_grad=True)
y = x + 2
# y was created as a result of an operation, so it has a grad_fn attribute.
# grad_fn: references a Function that has created the Tensor
print(x) # created by the user -> grad_fn is None
print(y)
print(y.grad_fn)
# Do more operations on y
z = y * y * 3
print(z)
z = z.mean()
print(z)
# Let's compute the gradients with backpropagation
# When we finish our computation we can call .backward() and have all the gradients computed automatically.
# The gradient for this tensor will be accumulated into .grad attribute.
# It is the partial derivate of the function w.r.t. the tensor
print(x.grad)
z.backward()
print(x.grad) # dz/dx
# !!! Careful!!! backward() accumulates the gradient for this tensor into .grad attribute.
# !!! We need to be careful during optimization !!! optimizer.zero_grad()
Stop a tensor from tracking history:
For example during the training loop when we want to update our weights, or after training during evaluation. These operations should not be part of the gradient computation. To prevent this, we can use:
x.requires_grad_(False)
x.detach()
- wrap in
with torch.no_grad():
# .requires_grad_(...) changes an existing flag in-place.
a = torch.randn(2, 2)
b = (a * a).sum()
print(a.requires_grad)
print(b.grad_fn)
a.requires_grad_(True)
b = (a * a).sum()
print(a.requires_grad)
print(b.grad_fn)
# .detach(): get a new Tensor with the same content but no gradient computation:
a = torch.randn(2, 2, requires_grad=True)
b = a.detach()
print(a.requires_grad)
print(b.requires_grad)
# wrap in 'with torch.no_grad():'
a = torch.randn(2, 2, requires_grad=True)
print(a.requires_grad)
with torch.no_grad():
b = a ** 2
print(b.requires_grad)
Gradient Descent Autograd
Linear Regression example:
$f(x) = w * x + b$
here : f(x) = 2 * x
import torch
# Linear regression
# f = w * x + b
# here : f = 2 * x
X = torch.tensor([1, 2, 3, 4, 5, 6, 7, 8], dtype=torch.float32)
Y = torch.tensor([2, 4, 6, 8, 10, 12, 14, 16], dtype=torch.float32)
w = torch.tensor(0.0, dtype=torch.float32, requires_grad=True)
# model output
def forward(x):
return w * x
# loss = MSE
def loss(y, y_pred):
return ((y_pred - y)**2).mean()
X_test = 5.0
print(f'Prediction before training: f({X_test}) = {forward(X_test).item():.3f}')
Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz
Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to ./data/MNIST/raw/train-images-idx3-ubyte.gz
100%|██████████| 9912422/9912422 [00:00<00:00, 116079447.08it/s]
Extracting ./data/MNIST/raw/train-images-idx3-ubyte.gz to ./data/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz
Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz to ./data/MNIST/raw/train-labels-idx1-ubyte.gz
100%|██████████| 28881/28881 [00:00<00:00, 120413214.54it/s]
Extracting ./data/MNIST/raw/train-labels-idx1-ubyte.gz to ./data/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz to ./data/MNIST/raw/t10k-images-idx3-ubyte.gz
100%|██████████| 1648877/1648877 [00:00<00:00, 28916457.87it/s]
Extracting ./data/MNIST/raw/t10k-images-idx3-ubyte.gz to ./data/MNIST/raw
Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz
Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to ./data/MNIST/raw/t10k-labels-idx1-ubyte.gz
100%|██████████| 4542/4542 [00:00<00:00, 5074727.96it/s]
Extracting ./data/MNIST/raw/t10k-labels-idx1-ubyte.gz to ./data/MNIST/raw
Top comments (0)