Introduction to Pytorch
Bikash Santra
Indian Statistical Institute, Kolkata
Author: Soumith Chintala¶
PyTorch¶
It’s a Python based scientific computing package targeted at two sets of audiences:
- A replacement for NumPy to use the power of GPUs
- a deep learning research platform that provides maximum flexibility and speed
Tensors¶
Tensors are similar to NumPy’s ndarrays, with the addition being that Tensors can also be used on a GPU to accelerate computing.
%matplotlib inline
from __future__ import print_function
import torch
1. Construct a 5x3 matrix, uninitialized:¶
x = torch.empty(5, 3)
print(x, x.dtype)
2. Construct a randomly initialized matrix:¶
x = torch.rand(5, 3)
print(x, x.dtype)
3. Construct a matrix filled zeros and of dtype long:¶
x = torch.zeros(5, 3, dtype=torch.long)
print(x)
4. Construct a tensor directly from data:¶
x = torch.tensor([5.5, 3])
print(x)
5. Help on 'torch' functions:¶
help(torch.transpose)
6. Transpose of a tensor:¶
x = torch.tensor([5.5, 3])
print(x, torch.transpose(x,0,0))
7. Creating tensor based on existing tensor:¶
These methods will reuse properties of the input tensor, e.g. dtype, unless new values are provided by user.
x = x.new_ones(5, 3, dtype=torch.double) # new_* methods take in sizes
print(x)
x = torch.randn_like(x, dtype=torch.float) # override dtype!
print(x) # result has the same size
8. Getting size of a tensor:¶
torch.Size is in fact a tuple, so it supports all tuple operations.
y = x.size()
print(y[0])
Operations on Tensors¶
There are multiple syntaxes for operations. In the following example, we will take a look at various operations.
1. Addition: syntax 1¶
y = torch.rand(5, 3)
print(x + y)
2. Addition: syntax 2¶
print(torch.add(x, y))
3. Addition: providing an output tensor as argument¶
result = torch.empty(5, 3)
torch.add(x, y, out=result)
print(result)
4. Addition: providing an output tensor as argument¶
# adds x to y
y.add_(x)
print(y)
Note
Any operation that mutates a tensor in-place is post-fixed with an ``_``. For example: ``x.copy_(y)``, ``x.t_()``, will change ``x``.
NumPy-like Indexing¶
print(x[:, 1])
Deep Copy of Tensors¶
a = torch.zeros((2,3))
print(a)
b = a.t().clone()
print(b)
b[0,0] = 5
print(b)
print(a)
Resizing / Reshaping Tensors torch.view:¶
x = torch.randn(4, 4, 4)
y = x.view(64)
z = x.view(-1, 2, 8) # the size -1 is inferred from other dimensions
print(x.size(), y.size(), z.size())
If you have a one element tensor, use .item() to get the value as a Python number¶
x = torch.randn(5)
print(x, type(x))
print(x.data[0], type(x.data[1]))
Read later:
100+ Tensor operations, including transposing, indexing, slicing,
mathematical operations, linear algebra, random numbers, etc.,
are described here <http://pytorch.org/docs/torch>.
NumPy Bridge¶
Converting a Torch Tensor to a NumPy array and vice versa is a breeze.
The Torch Tensor and NumPy array will share their underlying memory locations, and changing one will change the other.
a = torch.ones(5)
print(a, type(a))
1. Converting a Torch Tensor to a NumPy Array¶
b = a.numpy()
print(b, type(b))
See how the numpy array changed in value.
a.add_(10)
print(a)
print(b)
2. Converting NumPy Array to Torch Tensor¶
See how changing the np array changed the Torch Tensor automatically
import numpy as np
a = np.ones(5)
print(a)
b = torch.from_numpy(a)
np.add(a, 1, out=a)
print(b)
Note
All the Tensors on the CPU except a CharTensor support converting to NumPy and back.
CUDA Tensors¶
Tensors can be moved onto any device using the .to method.
# let us run this cell only if CUDA is available
# We will use ``torch.device`` objects to move tensors in and out of GPU
print(torch.cuda.is_available(), torch.cuda.get_device_name(0))
torch.cuda.empty_cache()
torch.cuda.set_device(0)
device = torch.device(0)
if torch.cuda.is_available():
y = torch.ones_like(x).to(device)
x = x.to(device) # or just use strings ``.to("cuda")``
z = x + y
print(z)
x = torch.rand(3,3)
y = torch.rand(3,3)
print(x)
print(y)
x = x.cuda()
print(x)
print(x+y)
y = y.cuda()
print(x+y)
From GPU to CPU
print(y.cpu())