. Pytorch Basics 2

[3]:

1

import numpy as np

[2]:

1

import torch

[4]:

12

# difference betwee tensor and Tensor 
arr=np.array([1,2,3,4,5])

[5]:

1

float_tensor=torch.Tensor(arr) # This is just a float tensor

[7]:

1

float_tensor # numbers will be float tpye

Out[7]:

tensor([1., 2., 3., 4., 5.])

[9]:

1

float_tensor.dtype # float type tensor

Out[9]:

torch.float32

[4]:

12

# Create a empty 2*2 tensor 
torch.empty(2,2)  # just empty block of memory is allocates , its not actully zeros

Out[4]:

tensor([[0., 0.], [0., 0.]])

[5]:

12

# to create actual zeros tensors 
torch.zeros(4,3,dtype=torch.int64)

Out[5]:

tensor([[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]])

[6]:

1

torch.ones(4,3) # for ones

Out[6]:

tensor([[1., 1., 1.], [1., 1., 1.], [1., 1., 1.], [1., 1., 1.]])

[7]:

1

torch.arange(0,18,2).reshape(3,3) #for range

Out[7]:

tensor([[ 0, 2, 4], [ 6, 8, 10], [12, 14, 16]])

[8]:

12

#linespace
torch.linspace(0,18,12).reshape(3,4)

Out[8]:

tensor([[ 0.0000, 1.6364, 3.2727, 4.9091], [ 6.5455, 8.1818, 9.8182, 11.4545], [13.0909, 14.7273, 16.3636, 18.0000]])

[9]:

1

torch.tensor([1,2,3]) # normal python list to tensor

Out[9]:

tensor([1, 2, 3])

[10]:

12

# change int to float , or increase precision use .type method 
my_tensor=torch.tensor([1,2,3])

[11]:

1

my_tensor.dtype

Out[11]:

torch.int64

[13]:

1

my_tensor=my_tensor.type(torch.float)

[15]:

1

my_tensor.dtype # 32 bit integer

Out[15]:

torch.float32

[17]:

123

# tensors with random numbers 
# random sample from uniform distribution(between 0 and 1)
torch.rand(4,3)

Out[17]:

tensor([[0.0642, 0.9729, 0.9474], [0.6554, 0.7476, 0.0173], [0.4503, 0.8049, 0.8815], [0.6116, 0.4514, 0.6972]])

[18]:

12

# normal distribution
torch.randn(4,3)

Out[18]:

tensor([[ 0.1052, -0.2184, -1.5320], [-0.8660, 1.6264, 0.3399], [-0.3530, -0.2078, -0.9368], [ 0.0481, -1.7387, -2.4484]])

[20]:

12

# random integers between range 
torch.randint(low=0,high=12,size=(5,5)) # you will not see 12 it is till 11

Out[20]:

tensor([[11, 2, 9, 3, 10], [ 1, 11, 3, 6, 1], [ 3, 8, 6, 0, 2], [11, 0, 7, 11, 0], [ 1, 3, 7, 10, 4]])

[21]:

123

# create random tensor given a size from some other tensor 
x=torch.zeros(2,5)

[22]:

1

torch.rand_like(x) # create shape from x and create random tensor , uniform distribution

Out[22]:

tensor([[0.9382, 0.5901, 0.5928, 0.5222, 0.9934], [0.1292, 0.5470, 0.3656, 0.2307, 0.7721]])

[23]:

1

torch.randn_like(x) # create shape from x and create random tensor , normal distribution

Out[23]:

tensor([[ 0.2755, 1.7455, -2.4249, -0.2167, 1.4576], [-0.4912, 0.7233, -1.3640, 1.2439, 1.9534]])

[26]:

1

torch.randint_like(x,low=0,high=12) # create shape from x and create random tensor , integer

Out[26]:

tensor([[ 9., 2., 0., 4., 11.], [11., 10., 9., 7., 6.]])

[34]:

123

# setting seed for random values 
torch.manual_seed(42)
torch.rand(2,3)  # seed should be in same cell , if different cell then it will produce different values

Out[34]:

tensor([[0.8823, 0.9150, 0.3829], [0.9593, 0.3904, 0.6009]])

[31]:

1

Out[31]:

tensor([[0.2666, 0.6274, 0.2696], [0.4414, 0.2969, 0.8317]])

[ ]:

1