Machine Learning

Question 1

What is Machine Learning?

Answer 1

Finds patterns in data then uses those patterns to predict the future

Question 2

What is the Learning?

Answer 2

Identifying patterns and then recognizing those patterns when you see them again

Question 3

What is the simple Machine Learning workflow?

Answer 3

• Feed data that contains patterns
• Machine learning algorithms find patterns
• This outputs a “model” that recognizes patterns
• Applications can use the model to get probabilities of match

Question 4

Good machine learning requires what?

Answer 4

• Lots of data
• Lots of computing power
• Effective algorithms

Question 5

Who cares about Machine Learning?

Answer 5

• Business leaders - want solutions to problems
• Software devs - create better applications
• Data Scientists

Question 6

What is a Data Scientist?

Answer 6

• Someone familiar with statistics
• Machine learning software (and ability to code it)
• Some problem domain (ideally)

Question 7

Who are some Machine Learning vendors?

Answer 7

• SAS Analytics
• RapidMiner Studio
• Alteryx Analytics

And “megavendors” and “cloud”
- IBM, SAP, Oracle, MS Azure, Amazon

Question 8

What is R?

Answer 8

• open source programming language and environment for machine learning
• very popular and many available packages
• been around a long time, since 90s
• most popular

Question 9

What are the 3 tenants of Machine Learning?

Answer 9

• Are you asking the right question?
• Do you have the data you need to answer that question?
• How do you measure success? How do you know when you are done?

Question 10

What is the Machine Learning pipeline?

Answer 10

• Raw data w/ someone w/ domain knowledge
• Pre-processing data into prepared data often many iterations until you are ready
• Apply learning alogorithm(s)
• Get Candidate Model and iterate to find best model
• Deploy the model
• Recreate model regularly based on new data and changing world

Question 11

What is Training Data?

Answer 11

• Prepared data used to create a model

- Creating a model is “training” a model

Question 12

What is Supervised vs Unsupervised Learning?

Answer 12

• Supervised - the value you want to predict is in the data, the data is “labeled”
• Un-supervised - the value you want is not in data, data is “not labeled”
• Supervised is most common

Question 13

What are Features in Machine Learning?

Answer 13

• “Features” are basically columns of the data

Question 14

What are categories of Machine Learning problems?

Answer 14

• Regression (supervised) - (how many will I sell next month)
• Classification (supervised) - (is CC trans fraud?) - returns probability
• Clustering (unsupervised) - what are our customer segments?

Question 15

What are types of algorithms?

Answer 15

• Decision tree
• Neural network
• Bayesian
• K-Means (for clustering)

Question 16

What are steps for training a model with Supervised Learning?

Answer 16

• Hire Data Scientist
• Choose features
• Send data to chosen algorithm
• Only send 75% of data
• Generate Candidate Model
• Test model with remaining 25% of data against candidate model