10 PTS
Due to Frank's bad rating as a passenger in an Uber, not many cars will pick him up when hailed. Frank has figured out that he has an 72% chance of not being picked up when he hails one of their cars. Out of the next 14 cars that Frank tries to hail, what is the probability that at least 10 will pick him up. Plot a bar chart that best illustrates the overall experiment.
5 PTS
You fit a linear regression to predict SAT score with many predictors, one of which is whether or not the student was homeschooled. Beta_homeschool = -40. How do you interpret the coefficient?
Being homeschooled reduces SAT by 40 points.
10 PTS
Give an example of a confusion matrix with accuracy > 90%, but both precision < 5% and recall < 5%. Prove your answer.
10 PTS
The playgolf.csv file has data on when is the best time to play golf. Imagine you're constructing a decision tree and the first split you make is on the Outlook column with the value Sunny. How much information have you gained by making that split?
5 PTS
You build a Decision Tree and get these stats:
train set accuracy: 90%
train set precision: 92%
train set recall: 87%
test set accuracy: 60%
test set precision: 65%
test set recall: 52%
What's going on? What tactic(s) do we have to modify our Decision Tree to fix the issue?
Tree is overfitting, start pruning.
5 PTS
You have a linear regression model that is overfitting on training data. What regularization techniques do you have to help with this issue? How do these techniques differ?
L1/L2
10 PTS
Implement the function column_averages. It takes the name of a csv file and returns a dictionary whose keys are the names of the columns and values are the mean of the column.
The file example.csv has been provided. The code should run as follows.
column_averages('data/example.csv')
{'A': 0.35999999999999999, 'C': 64.200000000000003, 'B': 1330.8}
10 PTS
I have two 6-sided dice. One is a standard die with the numbers 1-6. The other has three 3's and three 6's. You grab one at random and roll a six. What is the probability that you chose the fair die?
1/4
5 PTS
You are given the following data of student grades in a 5-by-5 matrix.
| id | Assignment 1 score | Assignment 2 score | Total score | Score percentage |
|---|---|---|---|---|
| 1 | 12 | 19 | 31 | 0.62 |
| 2 | 20 | 21 | 41 | 0.82 |
| 3 | 15 | 22 | 37 | 0.74 |
| 4 | 14 | 9 | 23 | 0.46 |
| 5 | 25 | 25 | 50 | 1.00 |
Make a guess of the rank of the matrix. Explain your guess.
3
5 PTS
You have a model that predicts the price of the house given the number of square feet, number of rooms, age and size of plot. You use a regression model to predict the price. You try varying the degree of the polynomial used to fit the data. This graph shows mean squared error on the training and testing sets based on the "complexity" of the model (degree of the polynomial).
What is going on in this graph? What would be an optimal choice for the degree of the polynomial?
4 to 6 is good
5 PTS
You have three models which classify emails as spam or not spam. You have tested it on a dataset of size 1000, with 100 spam emails and 900 non-spam emails. The accuracy of your models on that dataset is as follows:
Model A: 83%
Model B: 88%
Model C: 79%
Create a model which would have higher accuracy than any of the three models.
Predict all as not spam.
5 PTS
Logistic Regression internally creates a linear equation. How does the model use this linear equation to produce probabilities?
Use the sigmoid
5 PTS
What is grid search doing when it's running?
Creating different models with various hyperparameters, looking for the best model
5 PTS
If your linear regression model had a negative R-squared, what would be the simplest thing you could do to improve the model?
Predict the mean
5 PTS
Due to overfitting on your linear regression model, you decide to use L2 regularization. How do you know when lambda is tuned to the correct value for optimal model performance?
Look at your test data, when metrics are best
5 PTS
When scaling your data, why do you scale the test data set from the values in the training data set?
Test data represents future data.
5 PTS
When computing probabilities, the CDF of a binomial distribution uses sigma notation while the CDF of a gamma distribution uses integral notation. Why is this?
Binomial is discrete, Gamma is continuous.
5 PTS
Suppose you compare a five-predictor model with a higher R-squared to a one-predictor model. Is the five-predictor model better? Why?
Look at adjusted R-squared to verify
5 PTS
Why should you use relatively low dimensional vectors when using a KNN model?
Curse of dimensionality
10 PTS
At the manufacturing plant you work at, one empty 12-ounce aluminum soda should weigh about 15 grams. But you think the equipment has malfunctioned and is altering the weight of the cans. You perform a hypothesis test with the following data of measured can weight and a 97% confidence interval. What are your conclusions?
14.76515145, 15.66921114, 20.15497514, 14.25750049, 13.03577315,
8.26272148, 19.20985755, 12.04375879, 10.91085244, 11.68537558,
15.76500706, 9.10184183, 12.27624773, 12.54001656, 17.38484272,
17.57900201, 14.65171242, 15.81325345, 12.16072462, 9.66999714,
15.39782075, 14.71051375, 12.87376932, 13.16569868, 13.1352791 ,
17.62503595, 15.85801613, 16.08111466, 12.74932095, 8.78455273,
15.67603836, 20.85680148, 13.55898338, 16.12441324, 18.89939279,
9.74224957, 12.93047653, 15.26948315, 18.85026598, 9.6622572 ,
14.69180529, 7.98256903, 11.98738448, 13.09929852, 18.00647745,
13.37693129, 11.63440091, 19.55690723, 11.32272672, 9.65382463