Assume we are using regularized logistic regression for binary classification. Assume you have observed very large errors with its predictions on the new (unseen) data point, while the model has a low error on training data. Which of the following are steps would help to reduce the error on unseen data? Check all that apply. Use fewer training examples. Try adding polynomial features. O Try using a smaller set of features. Get more training examples.

You trained the regression model with 100 regressors and 1000 observations in the training and another 1000 in the test sample. You found that in-sample R2 over the training sample is 70% and the out-of-sample R2 over the test sample only - 30%. (select all that apply) a) Do you think there is any problem and how would you characterize it? Can adding more regressors (if you have them) help the model? b) Which approaches you may use to solve the problem? c) What would you expect the in-sample R2 to increase or decrease after that? What about the out-of-sample (test) R2?

Why use LASSO shrinkage methods for linear regression? Select ALL that are correct.     They allow for greater model interpretability.     They increase the prediction accuracy of linear regression on training data.     They slightly decrease bias and can substantially decrease the variance of the model.     They slightly increase bias, but can substantially decrease the variance of the model.

The benefits of switching to all-subsets regression from stepwise regression are broken forth in great depth below.   .

You are working on a spam classification system using regularized logistic regression. "Spam" is a positive class (y = 1)and "not spam" is the negative class (y=0). You have trained your classifier and there are m= 1000 examples in the cross-validation set. The chart of predicted class vs. actual class is: Predicted class: 1 Predicted class: 0 Actual class: 1 85 15 For reference: Accuracy = (true positives + true negatives)/(total examples) Precision = (true positives)/(true positives + false positives) Recall = (true positives)/ (true positives + false negatives) F1 score = (2* precision * recall)/(precision + recall) What is the classifier's F1 score (as a value from 0 to 1)? Write all steps Use the editor to format your answer Actual class: 0 890 10

We've built a logistic regression model in RapidMiner, and would like to use it to make predictions for some new data points. Which operator do we need: Performance Apply Model Cross Validation O Nominal to Numerical.

For evaluation of regression models, typically, the higher the      [ Select ]    ["Adjusted R Squared", "Residual Standard Error"]  the better, and the lower the      [ Select ]    ["Adjusted R Squared", "Residual Standard Error"]  , the better.

Question 48. Let us return to the Titanic data set. We now have learned several models and want to choose the best one. We used three different methods to validate these models: The training error rate (apparent error rate), the error rate on an external test set and the error rate estimated by a 10-fold cross validation. Training Error | Error on the test set | Cross Validation Error 0.18 Learner Decision Tree 0.22 0.21 Random Forest 0.01 0.10 0.12 1-Nearest-Neighbour 0.18 0.19 Which of the following statements are correct? a) 1-Nearest-Neighbour has a perfect training error and hence it should be used here. b) Random Forests outperforms both 1-Nearest-Neighbour and the Decision Tree in terms of prediction error. c) Not just in this case, but in general, Cross Validation is the better validation strategy and should always be preferred over the error on a single test set. d) Not just in this case, but in general, Decision Trees always perform worse than Random Forests.

This is a binary classification problem, y has two values (0 or 1), and X (feature) has three dimensions. • Use a logistic regression model to project X to y (classify X into two categories: 0 or 1). • The initialization is: w1 = 0, w2 = 0, w3 = 0, b = 0, Learning rate is 2. • You must use Gradient Descent for logistic regression in this question. • The regression should stop after one iteration.       Calculation process and formulas must be included in your answer! You must answer this question by manual calculation, but not programming.

Assume the following simple regression model, Y = β0 + β1X + ϵ ϵ ∼ N(0, σ^2 )   Now run the following R-code to generate values of σ^2 = sig2, β1 = beta1 and β0 = beta0. Simulate the parameters using the following codes:   Code:   # Simulation ##  set.seed("12345")  beta0 <- rnorm(1, mean = 0, sd = 1) ## The true beta0  beta1 <- runif(n = 1, min = 1, max = 3) ## The true beta1  sig2 <- rchisq(n = 1, df = 25) ## The true value of the error variance sigmaˆ2    ## Multiple simulation will require loops ##  nsample <- 10 ## Sample size n.sim <- 100 ## The number of simulations  sigX <- 0.2 ## The variances of X #   # Simulate the predictor variable ##  X <- rnorm(nsample, mean = 0, sd = sqrt(sigX))   Q1  Fix the sample size nsample = 10 . Here, the values of X are fixed. You just need to generate ϵ and Y . Execute 100 simulations (i.e., n.sim = 100). For each simulation, estimate the regression coefficients (β0, β1) and the error variance (σ 2 ). Calculate the mean of…