IEMS 5730讲解 、c++，Java语言编程辅导
IEMS 5730 Spring 2024 Homework 2
Release date: Feb 23, 2024
Due date: Mar 11, 2024 (Monday) 11:59:00 pm
We will discuss the solution soon after the deadline. No late homework will be accepted!
Every Student MUST include the following statement, together with his/her signature in the
submitted homework.
I declare that the assignment submitted on Elearning system is original
except for source material explicitly acknowledged, and that the same or
related material has not been previously submitted for another course. I
also acknowledge that I am aware of University policy and regulations on
honesty in academic work, and of the disciplinary guidelines and
procedures applicable to breaches of such policy and regulations, as
contained in the website
http://www.cuhk.edu.hk/policy/academichonesty/.
Signed (Student_________________________) Date:______________________________
Name_________________________________ SID_______________________________
Submission notice:
● Submit your homework via the elearning system.
● All students are required to submit this assignment.
General homework policies:
A student may discuss the problems with others. However, the work a student turns in must
be created COMPLETELY by oneself ALONE. A student may not share ANY written work or
pictures, nor may one copy answers from any source other than one’s own brain.
Each student MUST LIST on the homework paper the name of every person he/she has
discussed or worked with. If the answer includes content from any other source, the
student MUST STATE THE SOURCE. Failure to do so is cheating and will result in
sanctions. Copying answers from someone else is cheating even if one lists their name(s) on
the homework.
If there is information you need to solve a problem, but the information is not stated in the
problem, try to find the data somewhere. If you cannot find it, state what data you need,
make a reasonable estimate of its value, and justify any assumptions you make. You will be
graded not only on whether your answer is correct, but also on whether you have done an
intelligent analysis.
Submit your output, explanation, and your commands/ scripts in one SINGLE pdf file.
Q1 [20 marks + 5 Bonus marks]: Basic Operations of Pig
You are required to perform some simple analysis using Pig on the n-grams dataset of
Google books. An ‘n-gram’ is a phrase with n words. The dataset lists all n-grams present in
books from books.google.com along with some statistics.
In this question, you only use the Google books bigram (1-grams). Please go to Reference
[1] and [2] to download the two datasets. Each line in these two files has the following format
(TAB separated):
bigram year match_count volume_count
An example for 1-grams would be:
circumvallate 1978 335 91
circumvallate 1979 261 95
This means that in 1978(1979), the word "circumvallate" occurred 335(261) times overall,
from 91(95) distinct books.
(a) [Bonus 5 marks] Install Pig in your Hadoop cluster. You can reuse your Hadoop
cluster in IEMS 5730 HW#0 and refer to the following link to install Pig 0.17.0 over
the master node of your Hadoop cluster :
http://pig.apache.org/docs/r0.17.0/start.html#Pig+Setup
Submit the screenshot(s) of your installation process.
If you choose not to do the bonus question in (a), you can use any well-installed Hadoop
cluster, e.g., the IE DIC, or the Hadoop cluster provided by the Google Cloud/AWS [5, 6, 7]
to complete the following parts of the question:
(b) [5 marks] Upload these two files to HDFS and join them into one table.
(c) [5 marks] For each unique bigram, compute its average number of occurrences per
year. In the above example, the result is:
circumvallate (335 + 261) / 2 = 298
Notes: The denominator is the number of years in which that word has appeared.
Assume the data set contains all the 1-grams in the last 100 years, and the above
records are the only records for the word ‘circumvallate’. Then the average value is:
(335 + 261) / 2 = 298,
instead of
(335 + 261) / 100 = 5.96
(d) [10 marks] Output the 20 bigrams with the highest average number of occurrences
per year along with their corresponding average values sorted in descending order. If
multiple bigrams have the same average value, write down anyone you like (that is,
break ties as you wish).
You need to write a Pig script to perform this task and save the output into HDFS.
Hints:
● This problem is very similar to the word counting example shown in the lecture notes
of Pig. You can use the code there and just make some minor changes to perform
this task.
Q2 [20 marks + 5 bonus marks]: Basic Operations of Hive
In this question, you are asked to repeat Q1 using Hive and then compare the performance
between Hive and Pig.
(a) [Bonus 5 marks] Install Hive on top of your own Hadoop cluster. You can reuse your
Hadoop cluster in IEMS 5730 HW#0 and refer to the following link to install Hive
2.3.8 over the master node of your Hadoop cluster.
https://cwiki.apache.org/confluence/display/Hive/GettingStarted
Submit the screenshot(s) of your installation process.
If you choose not to do the bonus question in (a), you can use any well-installed Hadoop
cluster, e.g., the IE DIC, or the Hadoop cluster provided by the Google Cloud/AWS [5, 6, 7].
(b) [20 marks] Write a Hive script to perform exactly the same task as that of Q1 with
the same datasets stored in the HDFS. Rerun the Pig script in this cluster and
compare the performance between Pig and Hive in terms of overall run-time and
explain your observation.
Hints:
● Hive will store its tables on HDFS and those locations needs to be bootstrapped:
$ hdfs dfs -mkdir /tmp
$ hdfs dfs -mkdir /user/hive/warehouse
$ hdfs dfs -chmod g+w /tmp
$ hdfs dfs -chmod g+w /user/hive/warehouse
● While working with the interactive shell (or otherwise), you should first test on a small
subset of the data instead of the whole data set. Once your Hive commands/ scripts
work as desired, you can then run them up on the complete data set.
Q3 [30 marks + 10 Bonus marks]: Similar Users Detection in
the MovieLens Dataset using Pig
Similar user detection has drawn lots of attention in the machine learning field which is
aimed at grouping users with similar interests, behaviors, actions, or general patterns. In this
homework, you will implement a similar-users-detection algorithm for the online movie rating
system. Basically, users who rate similar scores for the same movies may have common
tastes or interests and be grouped as similar users.
To detect similar users, we need to calculate the similarity between each user pair. In this
homework, the similarity between a given pair of users (e.g. A and B) is measured as the
total number of movies both A and B have watched divided by the total number of
movies watched by either A or B. The following is the formal definition of similarity: Let
M(A) be the set of all the movies user A has watched. Then the similarity between user A
and user B is defined as:
………..(**) 𝑆𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦(𝐴, 𝐵) =
|𝑀(𝐴)∩𝑀(𝐵)|
|𝑀(𝐴)∪𝑀(𝐵)|
where |S| means the cardinality of set S.
(Note: if |𝑀(𝐴)∪𝑀(𝐵)| = 0, we set the similarity to be 0.)
The following figure illustrates the idea:
Two datasets [3][4] with different sizes are provided by MovieLens. Each user is represented
by its unique userID and each movie is represented by its unique movieID. The format of the
data set is as follows:
,
Write a program in Pig to detect the TOP K similar users for each user. You can use the
cluster you built for Q1 and Q2 or you can use the IE DIC or one provided by the Google
Cloud/AWS [5, 6, 7].
(a) [10 marks] For each pair of users in the dataset [3] and [4], output the number of
movies they have both watched.
For your homework submission, you need to submit i) the Pig script and ii) the
list of the 10 pairs of users having the largest number of movies watched by
both users in the pair within the corresponding dataset. The format of your
answer should be as follows:
, , //top 1
…
, , //top 10
(b) [20 marks] By modifying/ extending part of your codes in part (a), find the Top-K
(K=3) most similar users (as defined by Equation (**)) for every user in the datasets
[3], [4]. If multiple users have the same similarity, you can just pick any three of them.
Hint:
1. In part (b), to facilitate the computation of the similarity measure as
defined in (**), you can use the inclusion-exclusion principle, i.e.
|𝑆∪𝑇| = |𝑆| + |𝑇| − |𝑆∩𝑇|
In your submission, you only need to submit your Pig script together with the
following output:
i) the user in [3] whose ID shares the same last 2 digits of your CUHK student ID
number. For example, if your CUHK student ID number is ****13, you need to output
the Top-3 similar user list for User 013, 113, 213, 313, …, 513 (if such users exist in
the small dataset [3]), i.e.
013, , …,
113, , …,
...
513, , …,
ii) Similarly, for the dataset in [4], only output the Top-3 similar user list for the users
whose ID shares the same last 4 digits of your CUHK student ID number.
(c) [10 Bonus marks] Repeat part (b) using Hive.
Q4 [30 marks]: Spark Basic RDD
In this question, you will explore some of the basic RDD concepts in Spark. You need to
submit your Spark application to Hadoop cluster (either your own Hadoop cluster or
use the Spark Installation in DIC). You can use any programming language supported by
Spark but need to operate on the RDD level.
(a) [15 marks] Naive implementation of PageRank in Spark. In this part, write your own
PageRank program. Then, run your program on the given dataset [8], submit the
code and the top 100 nodes.
(b) [15 marks] Advanced implementation of PageRank in Spark. In this part, you need
to take advantage of the pre-partition mechanism to reduce the shuffling overheads.
Please adjust the number of partitions (try 3 different cases) and compare their
performance. Submit your result and explain your observations.
[Submission Requirements]
1. Submit the source codes and outputs of your programs in one single PDF report.
Besides, for each key step, you should also present the commands used (if
applicable), the descriptions (in pure texts), and illustrations (in figures/ screenshots)
that help convey and clarify your ideas for solving the problems.
2. Package all the source codes (as you included in Step 1) into a zip file.
3. Please submit both the PDF report and the zip file (i.e., you need to submit two
separate files) to CUHK Blackboard.
Reference:
[1] Google Books 1:
http://storage.googleapis.com/books/ngrams/books/googlebooks-eng-all-1gram-20120701-a.
gz
[2] Google Books 2:
http://storage.googleapis.com/books/ngrams/books/googlebooks-eng-all-1gram-20120701-b.
gz
[3] Small scale dataset
https://mobitec.ie.cuhk.edu.hk/ierg4330/static_files/assignments/movielens_small.csv
[4] Large scale dataset
https://mobitec.ie.cuhk.edu.hk/ierg4330/static_files/assignments/movielens_large_updated.c
sv
[5] Cloud Dataproc API
https://cloud.google.com/dataproc/docs/reference/rest
[6] Amazon EMR
https://aws.amazon.com/emr/
[7] One-click Hadoop-Cluster deployment using Amazon EMR (Zoom Recording, please
refer to Blackboard -> Course Contents -> Back-up Recordings for the passcode)
https://cuhk.zoom.us/rec/share/__ozUsXuqyHy3n5BA-MUm01bJedFK6IBfCqD6Qi2oCfcleG
J11rGTeqHi1_45hCO.xGoU3s3813Xn-_Q
[8] SNAP Google Web Data
https://snap.stanford.edu/data/web-Google.html