Data Science

Question 1

deep learning

Answer 1

subset of machine learning that can learn unsupervised from unstructured data (the deep refers to the depth of a networks layers)

Question 2

machine learning

Answer 2

subset of AI that automates analytical model building. emphasizes systems can learn from data, identify patterns, and make decisions

Question 3

artificial intelligence

Answer 3

theory and development of computer systems that normally require human intelligence

Question 4

3 types of machine learning

Answer 4

supervised learning, unsupervised learning, reinforcement learning

Question 5

data science

Answer 5

field of study that combines information technology, modeling, and business management to extract insights from data. Machine learning is a subset

Question 6

describe process of node in neural net deciding to fire?

Answer 6

takes the number passed from each of its connector nodes below it, multiplies each by its weight, and ]fires outgoing connections to nodes above if sum exceeds threshold val

Question 7

describe neural net training briefly

Answer 7

weights and thresholds are set to random vals. training data fed up from bottom layer (input layer) through successive layers (getting multiplied an added), until arrives transformed in output layers. Weights and thresholds continually adjusted until training data with same labels yield similar outputs

Question 8

Who first proposed neural nets?

Answer 8

McCullough and Pitts in 1944 at University of Chicago (later went to MIT to start first cognitive sci dept)

Question 9

neural net (and concerns)

Answer 9

means of doing machine learning, in which a computer learns to perform some task by analyzing training examples. Usually hand labelled in advance. Concerns are model transparency

Question 10

Rank decision tree, linear regression, random forest in descending interoperability, accuracy

Answer 10

1. Linear regression. 2. Decision Tree 3. Random forest

1. Random Forest 2. Decision Tree 3. Linear regression

Question 11

Examples of open source machine learning libraries

Answer 11

sklearn, tensorsflow, keras, H20

Question 12

evalML

Answer 12

featureLabs product that finds best model from many popular libraries to use

Question 13

featuretools

Answer 13

featureLabs product for constructing single table of features from multiple tables

Question 14

overfitting

Answer 14

capturing noise and patterns that don’t generalize well to unseen data (opposite is underfitting)

Question 15

supervised learning

Answer 15

uses labelled data and you have and training data with the “correct answer” you’re looking for

Question 16

unsupervised learning

Answer 16

uses unlabelled data to investigate patterns you’ve never considered

Question 17

decision surface

Answer 17

separates one class from another in a scatterplot

Question 18

Naive Bayes

Answer 18

common supervised classification algorithm.
Pros: easy, fast, handles tons of features (great for text classification)
Cons: doesn’t care about word order

Question 19

sensitivity

Answer 19

true positive rate from a test (or think if a test comes back positive, how likely is it to be true)?

Question 20

specificity

Answer 20

true negative rate from a test

Question 21

Bayes theorem

Answer 21

finds the probability of an event occurring given the probability of another event that has already occurred

Question 22

Symbol for ‘mean’ of x values

Answer 22

X-bar or X̄

Question 23

symbol for standard deviation of x

Answer 23

Sx

Question 24

correlation between x and y (symbol)

Answer 24

r

Question 25

number of values in a set (symobl)

Answer 25

n

Question 26

symbol for probability

Answer 26

P or p

Question 27

conditional probability symbol

Answer 27

pipe or | (can also indicate conditional distribution). Read usually as ‘given that’

Question 28

SVM

Answer 28

Support vector machine

Question 29

what does svm rely on?

Answer 29

a linear separation between the two classes (maximizes margin)

Question 30

how does svm compensate for non-linear space?

Answer 30

adds extra dimensions (called the kernel trick)