Homework answers / question archive / CSE 625 Parallel Programming Project 3 Readings:   [1] 04_C++_11_MultiThreading_1

CSE 625 Parallel Programming Project 3 Readings:   [1] 04_C++_11_MultiThreading_1

Computer Science

Share With

---

CSE 625 Parallel Programming Project 3

Readings:

 

[1] 04_C++_11_MultiThreading_1.pdf

[2] 04_C++_11_MultiThreading_2.pdf
[3] MNIST_2 All_pair_distance.pdf

 

CodeBlocks Project:

 

 All_Pair_Distance

 

Python Notebook and data:

            MNIST_Dataset.zip

 

Download link: https://louisville.box.com/s/g5ssgz9jgttxtd1u09o5xe2077gowfkj

 

It contains MINIST dataset

            train-images.bin            (60,000 28x28 float32 images)

            train-labels.bin                        (60,000 ubyte labels)

 

            test-images.bin              (10,000 28x28 float32 images)

            test-labels.bin              (10,000 ubyte labels)

 

and notebook, MNIST_Dataset.ipyn.

 

 

 

 

Assignment Description

 

1(40 points)

 

In the posted notebook, Mis_match.ipynb,  it defines a list called 

   which contains a list of 309 ordered pairs (test_idx, train_idx). For example, the first
   pair is (115, 8111) and it will be used to pair and compare 115-th test image with
   8111-th trainimage given in the MNIST dataset (i.e., test-images.bin and

   train-images.bin).

 

Based on this notebook, expand it to perform the following tasks:

 

1. Read the MNIST dataset into numpy arrays.
    
2. Plot 10 pairs in the list, , in a single plot. For each image in the plot, include a title of index of test or train image and its test or train label.
    
3. Take a user input of the index (between 0 and 308) of the pair in the list, mis_match, and plot the pair of images side-by-side with appropriate title (like that mentioned in task 2).

 

        Show a run of your notebook in the project report and submit your notebook to the
        Blackboard.

 

 

 

 

2 In the CodeBlocks project, All_Pair_distance, it implements four functions to
   compute the pair-wise distance matrix of MNIST train images (loaded from
   train-images.bin). These four methods are:

 

       1 sequential_all_pairs (sequential computing)
       2 block_all_pairs (C++ multi threads - block work distribution)

3 block_ cyclic_all_pairs (C++ multi threads - block cyclic work distribution)

       4 dynamic_all_pairs (C++ multi threads - dynamic work distribution)

 

2 .1 (22 points) Measure the runtimes of using these four functionss to compute the
      pair-wise distance matrix of various sizes and put the results in the following table:

 

Matrix Size	400	800	10,000	20,000	30,000	60,000
Method 1
Method 2 12 threads
Method 3 12 threads Chunk size 2
Method 4 12 threads Chunk size 2

 

 

1. 2. (8 points) In the report, explain the key ideas of the function, dynamic_all_pairs, of its work distribution implementation and how and why the std::mutex object is used.

 

 

3(15 points) In problem 2, suppose we use 3 threads in the function, block_all_pairs, to
   compute the pair-wise distance matrix of size 60,000-by-60,000 of the MNIST train
   images. Calculate the work amount (i.e., the number of outmost iterations) done by
   each thread. Show your work and the results. (Ref: MNIST_2 All_pair_distance.pdf)

 

 

4(15 points) In problem 2, suppose we use 3 threads and chunk size 1 in the function,
   block_cyclic_all_pairs, to compute the pair-wise distance matrix of size
   60,000-by-60,000 of the MNIST train images. Calculate the work amount (i.e., the
   number of outmost iterations) done by each thread. Show your work and the results.

   (Ref: MNIST_2 All_pair_distance.pdf)