3. Assume you have about 10K training examples. Which would you expect to take l

Question

3. Assume you have about 10K training examples. Which would you expect to take longer to train: a) a kNN classifier, or b) a logistic regression classifier? 4. (Yes/No) In fitting a logistic regression model, do we seek to minimize mean squared error, as we do with a linear regression model? 5. What would be a reason for using logistic regression rather than kNN for classification? (select all that apply) a. concern over available storage b. concern over the time needed to train c. concern over the time needed to make predictions

Explanation / Answer

3.The answer would be a)KNN classifier.The KNN classifier Is an Instance Based learner(IBL) and it is computationally incapable to classify large data sets. The most important parameter in KNN is K i.e. the nearest neighbour,the value of K is calculated as square root of N where N is the number of training samples.The kNN works well with numeric attributes but it is not able to work with nominal attributes,thus it is necessary to reduce the weight which is attached with variable and to reduce many variables at once,thus the best solution is to reduce the dataset. Whereas the Logistic Regression Classifier works with any kind and size of variables.

4.The Answer is NO because The prediction function in logistic regression is non linear.In Mean square error the prediction function needs to be squared which will result in non convex function along with local minimums,so if the cost function has many local minimums then the gradient descent will not find the optimal global minimum.

5. The answer is B and C

Related Questions

Navigate

• Browse (All)
• Browse 3
• Subjects

---

---

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.