8. Parameter Tuning n Bookmark this page You finally have your algorithms up and running, and a way to measure performance! But, it's still unclear what values the hyperparameters like T and should have. In this section, you'll tune these hyperparameters to maximize the performance of each model. One way to tune your hyperparameters for any given Machine Learning algorithm is to perform a grid search over all the possible combinations of values. If your hyperparameters can be any real number, you will need to limit the search to some finite set of possible values for each hyperparameter. For efficiency reasons, often you might want to tune one individual parameter, keeping all others constant, and then move onto the next one; Compared to a full grid search there are many fewer possible combinations to check, and this is what you'll be doing for the questions below. In main.py uncomment Problem 8 to run the staff-provided tuning algorithm from utils.py. For the purposes of this assignment, please try the following values for T:[1, 5, 10, 15, 25, 50] and the following values for [0.001, 0.01, 0.1, 1, 10]. For pegasos algorithm, first fix =0.01 to tune T, and then use the best T to tune , The most explanatory unigrams 10 points possible (graded) According to the largest weights (i.e. individual i values in your vector), you can find out which unigrams were the most impactful ones in predicting positive labels. Uncomment the relevant part in main.py to call utils.most_explanatory_word. Report the top ten most explanatory word features for positive classification below: Top 1: Top 2: Top 3: Top 4: Top 5: in 6 Top 6: Top 7: Top 8: Top 9: Top 10: Also experiment with finding unigrams that were the most impactful in predicting negative labels. Submit You have used 0 of 20 attempts Save