my-classnotes
Tensors and datasets
Quick overview of Tensors
Definitions:
- Tensor: a Pytorch tensor’s a data structure that’s generalization for numbers and dimensional arrays
1D Tensors
Tensors are ssentially arrays that’re the building blocks of a neural networks
Types of tensors given by dimention:
- 0-D: 1,2,0.2,10
- 1-D tensor is and array of numbers:
- A row in a database
- A vector
- Time series
A tensor contains elements of a single data type, the tensor type is the type of tensor:
Cheet cheat
import pytorch
a = torch.tensor[2,4,5,32,5]
b= torch.tensor([0.0,0.1,0.2])
#getting the tensor datatype
print(b.dtype)
#specify dtype
c = torch.tensor([0.0,0.1,0.2],dtype=torch.int32)
#explicity using dtype
d = torch.FloatTensor([1,2,3,4])
#converting tensor
a = a.type(torch.FloatTensor)
#getting the size and dimention of a tensor
e = torch.FloatTensor([1,2,3,4]
e.size() # torch.size(5)
e.ndimension() # 1
#converting 1-d to 2-d tensors using view method
f = torch.tensor([1,2,3,4])
f_col = f.view(5,1)
#if we didn't now the size
f_col = f.view(-1,1)
#pytorch tensors to numpy arrays
import numpy as np
np_array =np.array([0.0,0.1,0.2])
torch_tensor = torch.from_numpy(numpy_array)
back_to_numpy = torch_tensor.numpy()
#omg pointers again :o
import pandas as pd
#pandas series to tensor
pandas_series = pd.series([1,2,3,4,5])
pandas_to_torch = torch.from_numpy(pandas_series.values)
#converting tensor to a list
this_tensor = torch.tensor([0,1,2])
torch_tolist = this_tensor.tolist()
#indexing tensors
new_tensor = torch.tensor([5,3,42])
new_tensor[0] # tensor(5)
new_tensor[1] # tensor(3)
# using .item you can return a number
new_tensor[0].item()
Indexing and slicing
t = torch.tensor([5,3,42,54,2,3,5])
t[0] = 100
#slicing
d = c[1:5]
#replacing values by slicing
d[3:5] = torch.tensor([1,2])
Basic operations
#vector addition and substraction
#term by term
#has to be the same dtype
u = torch.tensor([1.0,0.0])
v = torch.tensor([0.0,1.0])
z = u+v # tensor([1,1])
#broadcasting
u = torch.tensor([1, 2, 3, -1])
v = u + 1 # [2,3,4,0]
#vector multiplication by scalar
y = torch.tensor([1,2])
x = 2*y
#product of two tensors
u = torch.tensor([1,2])
v = torch.tensor([3,2])
z = u*v # [3,4]
#dot product uT*V
u = torch.tensor([1,2])
v = torch.tensor([3,2])
result = torch.dot(u,v) #5
Appliying functions to tensors
#mean
a = torch.tensor([1,-1,1,1])
a.mean()
#max
a = torch.tensor([1,-1,1,1])
a.max()
#map
np.pi
x = torch.tensor([0,np.pi/2,np.pi])
y = torch.sin(x) # [0,1,0]
#linsoace
torch.linspace(-2,2,steps=5)
#[-2,-1,0,1,2]
torch.linspace(-2,2,num=9)
#[-2,-1.5,-1,-0.5,0,0.5,1,1.5,2]
#plotting math functions
x = torch.linspace(0,2*np.pi,100)
y = torch.sin(x)
import matplotlib.pyplot as plt
#convert to numpy array
plt.plot(x.numpy(),y.numpy())