2016 Pittsburgh Data Jam Student Workshop

Pittsburgh Data Jam 2016
Bringing Big Data Education and Awareness to
Pittsburgh High School Students
February 26, 2016

Introductions
Saman Haqqi - President - Pittsburgh Dataworks
 saman.haqqi@pghdataworks.org
Brian Macdonald – Data Scientist – Oracle Corporation
 brian.macdonald@oracle.com
Pitt Science Outreach
 Margaret Farrell mef85@pitt.edu
 Laura Marshall LJM82@pitt.edu
 Jenny Lundahl jal225@pitt.edu
 Jackie Choffo jac335@pitt.edu
 Kyle Wiche KAW196@pitt.edu
 Chris Davis CJD81@pit.edu

Mentors
 Each team will be assigned a mentor
Can ask questions via email at any time
 Copy everyone on your team
 Copy your teacher
Pitt Science Outreach students
 Send email to all
Have a regular scheduled call with your mentor
 Don’t wait to right before presentations.

Data Analysis Workshop
Today’s Goals
Identifying relevant variables
Depicting them graphically
Doing the analysis
Drawing conclusions
Making recommendations

What technology will you use?
Lots of tools are available
Keep it simple at the beginning
Use Excel
Tableau is also available
Many Others
 R, SAS, Cognos, Oracle Business Intelligence, Google Apps,
Matlab, Pyhton, Spotfire, QlikView

Data Analysis Process
A standard repeatable process to guide data analysis.
Used formally and informally
 If you do analysis, you will do these steps.
Used for Big Data or not so Big Data
Becomes second nature as you do more analysis.
Is not about using a cool data analysis tool
 Although they are extremely helpful.

The Data Analysis Process
Define your Problem
Identify Data
Plan your Analysis
 Explore Data
 Prepare Data
 Model Data
Tell A Story
Make Recommendations
Determine What’s Next
Today’s Focus
In practice it looks like this
https://cyberitgs.wikispaces.com/Sandbox+Yerlan

Basic Steps for Analysis
Data Exploration
Data Preparation
Build Models

Data Exploration
Exploratory Data Analysis (EDA)
 Goal is to get an understanding of what data you have
What are your variables
Basic Statistics
Graph Data
Look for missing values
Look for outliers
Will this data help you answer your question?

Basic Statistics
Goal is to get a basic understanding of your data
 Mean (Average)
• Sum of values/Count of values
 Median
• Mid Point of Values
 Maximum, Minimum (Range)
 Standard Deviation (σ) & Variance (σ^2)
• How spread out the values are compared to the mean
 Quartiles
• Nice buckets of the spread of the data

Graphing Data
Helps visualize patterns in the data
Especially with large data sets.
 https://www.mapbox.com/labs/twitter-
gnip/locals/#12/40.4620/-80.0151
Spot exceptions
Use the best graph for the data types
Help tell your story

Missing Values
Can have large impact on basic statistics
Count # of missing values of every variable (column)
Important to understand why data is missing?
 Data entry
 Wasn’t collected
 Isn’t relevant
Should you use the variable?
Should you fill in missing values
 Use mean, median, max, min, 0.
 You need to determine best method

Outliers
Outliers are values at the extreme
Much larger or smaller than most of your data
May have many causes
 Data Entry Error
 Instrument Malfunction
 Real Exceptional data
Is 140º F an Outlier
Some are easy to spot within a single variable
Some are only found with multiple variables

Outliers
Need to decide how to treat Outliers
 Is the variable ok to use? Do you question the validity of the
data?
 Remove them from your data set?
 Keep them as is?
 Change the value (i.e. make it less extreme)
 Infer the real meaning
• -90º F temperature in Miami is likely 90º
Make sure you understand implications
Document your decision making

Demo – Missing Values &
Outlier Detection in Excel

One Last Thought on Exploring Data
You must be observant
Count the Number of F’s in the following sentence.
 You will have 15 Seconds
FINISHED FILES ARE THE RE-
SULT OF YEARS OF SCIENTIF-
IC STUDY COMBINED WITH
THE EXPERIENCE OF YEARS.

Leave your assumptions at the door!
FINISHED FILES ARE THE RE-
SULT OF YEARS OF SCIENTIF-
IC STUDY COMBINED WITH
THE EXPERIENCE OF YEARS.

Exploration Exercise
Using Excel
Sort
Filter
Summarize
Create Crosstabs
Charting

Data Preparation
 This step will fix any issues you found during data exploration
 Fix missing values
 Remove bad data
 Create new variables
 Add/Subtract/Multiply/Divide multiple variables
 Ratios
 Binning
 Other functions like Square Root or Exponents
Anything else you feel appropriate
 Have fun and experiment. You can not hurt data.

Preparation Exercise
Using Excel
Merge data
New Calculations
Fix Missing Data
Fix Outliers

Explaining Insights
How do you know what you
see is valid?
And not due to chance?
Correlation
http://musicthatmakesyoudumb.virgil.gr/

Correlation
The degree to which two or more attributes or measurements on the
same group of elements show a tendency to vary together
Positive when values increase together
Negative when values decrease together
http://www.mathsisfun.com/data/correlation.html

What can you tell me about this graph?
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80
Ice Cream Consumption/Capita
Ice Cream
Consumption/Capita
Linear (Ice Cream
Consumption/Capita)
IceCreamconsumption/capita
Drownings

Does Ice Cream Consumption Cause
Drowning?
Obviously not
Correlation does not imply Causation
 One may cause the other, but correlation just defines how
they vary.
 There may be other reasons. i.e. Hot temperatures
Be very cautious with Causation
 There are tests to determine causation

How do I know if variables are correlated
R = Correlation Coefficient
 Values between -1 & 1
 Positive Correlation > 0 - As one variable increases, the other
increases
 Perfect Correlation = 1
 Negative Correlation < 0 - As one variable increases, the other
decreases
 Perfect Negative Correlation = -1
 0 = No correlation
 Can be shown with a trend line
Understanding R and R2

How do I know if variables are correlated
R2 = Coefficient of Determination
 Tells how likely one variable predicts the other variable
 Values between 0 & 1
 If R 2 = 0.850, 85% of the total variation in y can be explained
by the linear relationship between x and y
 R2 is more commonly used
Understanding R and R2

Some Terminology
Independent Variable
 These are the variables that you modify
 In trend equation they are the X values
Dependent Variable
 These values depend on the values of the Independent
variables.
 In trend equation they are the Y values
y = 0.0045x + 691.18
y is Living Area
x is Sale Price
Slope Intercept

Modeling Exercise
Using Excel
Create scatter plot
Show Coefficient of determination
Create a formula to predict a value

What did the Data Tell You
Did it support your initial question?
 What conclusions can you make?
 Make sure they are fact based
 Check your bias
What is your story?
 Is it compelling?
• Does x influence y?
 Can it support actions to be taken?
 If not, is there still some benefit?

What did the Data Tell You
What recommendations will you make?
 Will you stand behind them?
 If not, why not?
 Can they really be implemented?
 What is the value of implementing the recommendation
What new questions would you ask?
 To clarify your analysis?
 Expand on your analysis
 Can better questions be asked?

And the most important Item
Have

Questions?
Always ask questions!!!!

Timing
Introductions – 10 Minutes
Overview/Data exploration Lecture – 35 Minutes
Exploration Hands-on – 30 Minutes
Data Prep Lecture – 20 Minutes
Data Prep Hands-on – 25 Minutes
Data Modeling Lecture – 20 Minutes
Data Modeling – Hand-on – 30 Minutes
Questions/Wrap Up – 10 Minutes
Total 3:00

2016 Pittsburgh Data Jam Student Workshop

More Related Content

What's hot (19)

Viewers also liked (11)

Similar to 2016 Pittsburgh Data Jam Student Workshop (20)

Recently uploaded (20)

2016 Pittsburgh Data Jam Student Workshop