Project-Part2 Deadline: April. 20th

[HuyTu7]

Building on Part 1 of the project, and your understanding of the classifiers you have implemented, you are to demonstrate your understanding by validating the choice of hyper-parameters of each algorithm and its associated classification performance.

1. Cross-Validation In order to find the best performance as well as most generalized model for your algorithm, a 5-fold cross validation is to be applied in your training phase. Explain how this Cross-Validation helps you find your hyperparameters.

2. Demonstration of Performances a. A “demonstration” implementation of the algorithm should be run as a command to train the classifier, test the classifier, print the performance on training and testing datasets b. Draw the following graphs: i) convergence, ii) performance vs. time, iii)performance vs. different hyperparameters.

3. Analysis of Results a. Based on the performance evaluation in Q.2, analyze the results by discussing the advantages as well as drawbacks of each algorithm, and how they may arise b. To avoid such drawbacks, how would you go about improving an overall smooth performance?

[HuyTu7]

1.

• add more params?
• run cv for 5 folds + sci-kit learn to pick the best param set
• run cv for 5 folds for the best param set and the default params 3.
• run statistical test (scottknott test to see if the results are statistically significant from each other)