Advanced Data Science

Homework 3

25 points

1. Use ssh to connect to bl2.cs.fsu.edu and login with your bl2 user name
you used for Homework 1. You can do it either from the internal network
(any machine in cs.fsu.edu domain) or outside (e.g. from shell.cs.fsu.edu)
2. Consider file traindata.csv outside your home folder (../traindata.csv). It
contains synthetic labeled data in a form of a table, each row having a
patient who did or did not have a heart attack in the past. It is a synthetic
dataset composed for this assignment, it does not have any real medical
data.
a. Each row has more than 10 attributes and the last attribute
(Result) – a label (yes/no), indicating whether a patient did or did
not have a heart attack in the past.
3. Separate a part of the file traindata.csv (for example 1/15th) and save it as
a separate file - testdata.csv. Remove this 1/15th from traindata.csv and
do not use it for step #4.
4. Browse the file (traindata.csv) to get familiar with its contents and
manually design a formula that can calculate the label value, based on
the attributes values.
IMPORTANT!!:
1. DO NOT USE ANY EXISTING AI MODELS (e.g. Decision Trees,
Linear/Logistic Regression, SVM, Naïve Bayes, or any other). DO
NOT DEVELOP ANY AUTOMATIC ALGORITHMS OR MODELS BY
YOURSELF.
2. DO NOT CALCULATE CORRELATION BETWEEN FEATURES OR
THE LABEL. DO NOT USE ANY MATH PACKAGES SUCH AS R,
MATLAB, ETC TO AUTOMATE THE FEATURE ENGINEERING OR
DEVELOPMENT PROCESS.
 If you do (let’s say take and train an already existing Decision Tree or any
other model then take the automatically generated rules and use them to
help you solve this assignment OR calculate correlation between features
and the label and use it for feature selection), then you’ll lose all credit for
this assignment.
a. Start from something simple, for example:
i. if (Age>=55 and Gender=1) then Result=no
else if (Age<39 and Gender=0) Result=yes;
Note that this is not the final rule that you need to turn in, but just
an example of how the design process can be started.
b. As you start designing and trying out your rule, you will see it
works for some rows (i.e. the Result generated by your rule is the
same as the Result value already present in the row). Try
changing or enriching your rules with more attributes, so the final
rule performs better (i.e. makes less mistakes compared to the
Result value present in the rows).
c. Make iterations to optimize your rule to make it perform as
accurate as possible.
d. You can use any suitable programming language to execute your
rule on the test set.
5. When you finish changing your rule, calculate its accuracy on testset.csv
(see step #3) by using the following formula:
#𝑟𝑜𝑤𝑠 𝑖𝑛 𝑡ℎ𝑒 𝑡𝑒𝑠𝑡 𝑠𝑒𝑡−#𝑖𝑛𝑐𝑜𝑟𝑟𝑒𝑐𝑡 𝑙𝑎𝑏𝑒𝑙𝑠 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑒𝑑 𝑏𝑦 𝑦𝑜𝑢𝑟 𝑟𝑢𝑙𝑒 𝑜𝑛 𝑡ℎ𝑒 𝑡𝑒𝑠𝑡 𝑠𝑒𝑡
a. 100% ∗ #𝑟𝑜𝑤𝑠 𝑖𝑛 𝑡ℎ𝑒 𝑡𝑒𝑠𝑡 𝑠𝑒𝑡

What to turn-in
Turn in a PDF file having the following items:

1. (15 points) The rule you designed to label the data rows
2. (5 points) Explanation of your rule and the intuition that you have used to
design it
3. (5 points) The accuracy of your rule calculated on testdata.csv separated
from the main file – traindata.csv in step #3.
4. Do not forget to write your name at the beginning of the file.
Good luck!