Day #221 - Session 5-8 - Pytorch

Key Lessons
Session #4

• Tensor Attributes
• Inputs / Outputs represented by tensors
• Scalar, Vector, Matrix (Used in maths)
• Number (Zero Index), Array (One Index), 2D-Array (Two Index)
• Tensors are multi-dimensional arrays
• Tensor Attributes, Rank, Axes, Shape

Session #5 

• Rank - Refers to the number of dimensions present in tensor, How many indices required to access an element
• Shape - Length of each axes, Shape allows visualizing tensor

Session #6

• Image Input - to CNN as tensor
• Axes from Right to left
• [? A0,? A1,? A2,? A3] 
• Height / Width on last axes
• [?,?,H,W]
• Color Channels
• [?,C,H,W]
• Three Indexes for Color Channel
• First Axes - Batch Size (Batches of samples)
• [3,1,28,80
• Batch of 3 images
• Single Color Channel
• Height, Width 28,28
• Three channel output for three layers

Session #7

• Data to Tensors
• torch.tensor class
• Top level torch package
• Tensors contain data of uniform type
• Computation depend on type and device
• Computation has to be with same datatypes and same devices
• matching CPU and GPU versions

	import torch
	import numpy as np
	t = torch.Tensor()
	type(t)
	#type of data contained in tensor
	print(t.dtype)
	print(t.device)
	print(t.layout)
	#strided
	#data loaded in memory
	data = np.array([1,2,3])
	type(data)
	#specify cpu or gpu
	device = torch.device('cuda:0')
	data = np.array([1,2,3])
	torch.Tensor(data)
	torch.as_tensor(data)
	torch.from_numpy(data)
	#Identity matrix
	torch.eye(2)
	#zeros functions
	#Rank 2 tensor
	torch.zeros(2,2)
	torch.ones(2,2)
	torch.rand(2,2)

view raw ptrotch1.py hosted with ❤ by GitHub

Happy Mastering DL!!!