ML_HW2_Writeup_final

pdf

School

Northeastern University *

*We aren’t endorsed by this school

Course

5644

Subject

Computer Science

Date

Feb 20, 2024

Type

pdf

Pages

21

Uploaded by Codja

Report
EECE 5644 Introduction to Machine learning and Pattern Recognition Homework 2 Alvin Joseph NU ID: 002197786 1. Part 1: Class distribution and true class labels scatter graph for 20 samples trained
Class distribution and true class labels scatter graph for 200 samples trained Class distribution and true class labels scatter graph for 2000 samples trained
Class distribution and true class labels scatter graph for 10000 samples trained The classifier was implemented on the samples generated by the dataset D 10000 validate and the below ROC was obtained.
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
Observed Results : Experimental Gamma value: 1.4770321978241072 Theoretical Gamma value: 1.85 Experimental minimum error: 0.16508329082792625 Theoretical Minimum Error: 0.2120656046245883 Difference of 0.047 observed between experimental and theoretical minimum error. Given below is the supplementary visualization plot generated which indicates the decision boundary based on the classification rule of the dataset.
Part 2 (a) Maximum likelihood parameter estimation was used to train the logistic linear function on the basis of class label posterior function on the D 20 train, D 200 train and D 2000 train datasets.
Training Dataset Accuracy Min probability of error D 20 train 0.5887 0.4113 D 200 train 0.6262 0.3738 D 2000 train 0.6241 0.3759 (b) Maximum likelihood parameter estimation technique to train a logistic-quadratic-function based approximation of class label posterior function on the D 20 train, D 200 train and D 2000 train datasets
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
Training Dataset Accuracy Min probability of error D 20 train 0.7652 0.2348 D 200 train 0.8329 0.1671 D 2000 train 0.8498 0.1508 Thus, we can see that logistic quadratic function yielded better results than its linear counterpart based on the values obtained. Logistic quadratic function also provides better results than the theoretical optimal classifier in part 1.
2. 100 samples 1000 samples
Maximum Likelihood Estimator (MLE) MLE Error: 4.723921802780472 Maximum A Posteriori Estimator (MAP)
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
Gamma values Error 10 -4 4.7239218918550545 10 -3 4.723922693595445 10 -2 4.723930717861503 10 -1 4.724011631168565 1 4.7248740146759935 10 4.73315967349211 10 2 4.732939362650383 10 3 4.739204050106735 10 4 4.777870052766509 Not much difference was observed between error values of the ML estimator and the MAP estimator for the corresponding gamma values. As the gamma values increases by a factor of 10 the error also increases by a small fraction.
Appendix: 1.
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
2.
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help
References: 1. Github 2. Professor Mark s codes 3. https://arxiv.org/ftp/arxiv/papers/1811/1811.01043.pdf 4. http://www.cs.columbia.edu/~mcollins/em.pdf 5. Murphy s book
Your preview ends here
Eager to read complete document? Join bartleby learn and gain access to the full version
  • Access to all documents
  • Unlimited textbook solutions
  • 24/7 expert homework help