Show:
Compare Those Models!
Analyze This (Network)
Clusterin'
Using Your Neurons
Potent Potables
100
When comparing models using confusion matrix metrics (for example, accuracy or sensitivity), which data set should we use to calculate those metrics: training, validation, or testing?
Testing
100
If a node has an indegree of 3 and an outdegree of 5, what is its degree?
8 (3+5)
100
True or False: the "k" in k-means clustering is the same as the "k" in k-nearest neighbors?
False!
100
A transfer (or activation) function always returns a value between 0 and what other number?
1
100
Ninety-nine bottles of beer on the wall, ninety-nine bottles of beer, take one down, pass it around...and be thankful calculating naive what isn't on the final?
Naive Bayes
200
Like adjusted R2 for linear regression, AICc for logistic regression accounts for the number of what in your model?
Variables
200
True or False: a network with high connectivity is always preferred over a network with low connectivity.
False!
200
Because clustering (usually) deals with problems where we have no response/Y value to predict, we call this what type of learning?
Unsupervised
200
Neural network methodology gets its name and its model from what part of the body?
The brain
200
If I split my wine quality data set into equal 20% validation data sets, how many times will I run my method under cross-validation?
5
300
Logistic regression, k-means clustering, hierarchical clustering, neural networks, naive Bayes: which one is the least interpretable?
Neural networks
300
Six Degrees of Kevin Bacon (or Monica Lewinsky) is an example of what kind of network?
A small world network
300
What is the main disadvantage to using K-Means Clustering to evaluate a data set?
You must specify the number of clusters (k)
300
Common transfer (or activation) functions include thresholding, Gaussian/normal, linear, and what other type of function used as the default in both JMP and Radiant?
Sigmoid (S-curve)
300
I run a regression to predict whether vodka drinkers prefer straight vodka over flavored vodka using multiple variables, including Age. If the coefficient for Age is 1.67, what needs to be added to the following interpretation for it to be an accurate statement? "For every one additional year of age, the odds of preferring straight vodka over flavored vodka increase by a factor of e1.67."
"...keeping all other variables fixed."
400
Put the following three methods in order from (generally) best for prediction to worst for prediction: hierarchical clustering, k-nearest neighbors, neural networks.
Neural networks (best), KNN, hierarchical clustering
400
If Nodes A, B, and C are strongly connected, what is their clustering coefficient?
1
400
Cluster 1 has two points, A and B. Cluster 2 has two points, C and D. When using the single linkage method, we record the distance between A and C. What must be true about A and C?
They are the closest points (between the two clusters)
400
If a neural network has 5 inputs, 2 nodes, and 1 output, how many total weights/estimates will it produce (not counting intercepts)?
10 (5*2) input-to-node connections plus 2 node-to-output connections (12 total)
400
When running a classification tree to predict if someone prefers red wine over white wine, you notice the first split is made on Gender; the next split on both sides is made on Income. In a logistic regression for the same problem, which of the variables will be significant?
Unclear--tree splits don't guarantee significance!
500
What single metric can you use to compare across any models, regardless of their methodology?
None!
500
In a random network with 51 nodes and a probability of connection of 0.2, how many connections is Node 1 expected to have?
10 (50 other nodes * 0.2 chance of connecting to each node)
500
If you cut a dendrogram as close to the bottom as possible, how many clusters will you have?
n (as many clusters as there are points)
500
There are three general parts to a neural network: the neural network graph to define the network structure, the recall techniques to determine how information is obtained from the network, and what third part?
Learning algorithm (how learning takes place in the network)
500
Rounding to the nearest ten dollars, what will your 431 professor's bar tab be on Thursday, May 18th?
$0--an analyst never pays bar prices when she can forecast ahead of time!