CS973-HW2
Q1 Perceptron Learning
10 Points
(i) Explain in your own words within 2-3 sentences the concept of perceptron learning?
(ii) Does perceptron learning always converge? If not, what property must the data have on which perceptron learning is guaranteed to converge?
(iii) What are the hyper parameters of perceptron learning? How would you decide on the value of the hyper parameters?
Save Answer
Q2 Weights in Perceptron Learning
10 Points
(i) Explain in your own words with simple arguments that if during the training, the activation output for data sample i does not match the class label for data sample i, then for data sample i+1, the new weight for feature j will be updated proportional to the feature value j of the ith sample.
(ii) Download the example code from https://github.com/rasbt/python-machine-learningbook/tree/master/code/ch02. Experiment by initializing the initial weights to zero rather than small random values. Explain your observations as to what differences you find compared to the case when the weights are initialized with random values.
Save Answer
Q3 Scikitlearn Python Functions
10 Points
In the Python object oriented programming model, a machine learning model is encapsulated in an object. The object is initialized with hyperparameters. The object is fed training data with a function fit(), and to make a prediction on test data, a function, predict() is used. Also, to split the given data into training data and test data set, a function train_test_split() is used. Do some research on these functions, and explain the input, output parameters of these functions, and their roles in the functions.
Please make sure that you write the explanations on your own, and do not cut-and-paste from the Internet or Scikitlearn manual.
Save Answer
Q4 Adaline Algorithm
10 Points
(i) What are the two most significant differences between the perceptron learning algorithm and Adaline Gradient Descent algorithm. (Please think carefully — what are the most significant differences?
(ii) Explain why a wrong learning rate can throw gradient descent into a non-convergent path?
Save Answer
Q5 Adaline Stochastic Gradient Descent
10 Points
Download the code from https://github.com/rasbt/python-machine-learning-book/tree/master/code/ch02
Study the code, and explain the need for the partial_fit() function in the AdalineSGD object?
Run the AdalineGD and AdalineSGD algorithms in the downloaded code, and then explain your observations with respect to the behavior of the two algorithms.
Save Answer
Save All Answers Submit & View S