Description

Lab 2: Linear Algebra and basic Torch
Lecturer: Abir De
Note: Please read the instructions mentioned in the questions carefully. We have provided boilerplate code for each question. Please ensure that you make changes in the areas marked with TODO.
1 LU Decomposition
LU decomposition refers to the factorization of square matrix A into two factors, a unit in lower triangular matrix L and an upper triangular matrix U, A = LU.
Your task in this question is to obtain the matrices L and U for a given square matrix A ∈Rn×n.
Complete the function LU_decomposition in the script.
Function Signature: def LU_decomposition(A: np.array):
This returns two matrices L,U which are of same shape as A.
2 Simple Torch
Using the PyTorch library, write Python scripts to achieve the tasks listed below. To install the PyTorch library, you can use the following command:
pip3 install torch –extra-index-url https://download.pytorch.org
/whl/cpu
Tasks:
1. Create and return a torch matrix A of shape 50×40×5 containing random numbers in the range [0,1).
2. You are given a matrix B of datatype float32. Return this matrix after converting it’s datatype to int32
3. Generate a random matrix C of shape 3×100. Permute the rows of the matrix to generate a new matrix D, such that the 1st row becomes the 2nd row, the 2nd row becomes the 3rd row and the 3rd row becomes the 1st. Return both C and D.
4. Generate a random matrix E of shape 20 × 10. Compute the sum along each row of this matrix in a new vector F. Return both E and F
5. You are given three matrices, G1,G2,G3, each of shape 10×10. Generate a new matrix H of shape 10×10×3 by combining the given matrices. Return H
3 Vectorization
In this problem, we will implement a function to compute the pairwise ranking loss between two sets of scores. Given a set of positive scores P ∈Rn1×1 and a set of negative scores N ∈Rn2×1, we want to impose the constraint that the positive scores should be greater than the negative scores. Therefore, we design the following loss function:
L(P,N) = X max(0,n − p)
p∈P,n∈N
1. Complete the function pairwise_ranking_loss_looped using for loops to obtain L(P,N) for any given vectors P and N.
2. Note that vectorized operations can make the above computations blazingly fast. Complete the function pairwise_ranking_loss_vec to obtain the same value as above in a vectorized manner.
We will be checking the execution time of the functions rigorously, so make sure that pairwise_ranking_loss_vec has no for loop. Also make sure that the two functions pairwise_ranking_loss_looped and pairwise_ranking_loss_vec, return the same value for any given input. The main function is provided. You can manipulate the scores P, N to observe the difference between the execution times of the functions.
Complete the functions in assignment.py. Keep the file in a folder named <ROLL_NUMBER>_L1 and compress it to a tar file named <ROLL_NUMBER>_L1.tar.gz using the command
tar -zcvf <ROLL_NUMBER>_L1.tar.gz <ROLL_NUMBER>_L1
Submit the tar file on Moodle. The directory structure should be –
<ROLL_NUMBER>_L1
| – – – – assignment.py
Replace ROLL_NUMBER with your own roll number. If your Roll number has alphabets, they should be in “small” letters.