SEDS-HW4

Homework #4 from SEDS

note This is almost certainly going to be the most challenging homework so far. It will build on everything you have learned from flow control, functions, lists, etc. Although these instructions put the functions before the unit tests, you should follow the test driven development paradigm.

ALL CODE in .py files should be pep8 compliant.

Assignment

1. Create a .py file called knn.py that contains your own implementation of a k-NN classifier. Hint You should have the following functions (at least): (2 pts)

• A wrapping function that is the primary way of interacting with your code. It takes as parameters, a training dataframe, a value of k and some input data to be classified. It returns the classification for the input data.
• A function that returns the Euclidean distance between a row in the intput data to be classified.
• A function that returns the list of sorted Euclidean distances between the input data and all rows in the dataframe. Hint Append the distances associated with the rows to a list and use the .sort() method on your list.
• A function that returns the class prediction based on the list of sorted Euclidean distances.
• A wrapping function that helps the user decide on what k to use. This function takes as parameters, a training dataframe, a testing dataframe and a list of values of k to try. It returns a dictionary with k as the keys and the training accuracy of the test set. Accuracy is measured by percentage of classifications that were correct for that value of k.

2. Create a new Jupyter notebook called 'SEDS-HW4.ipynb' that documents how to use your k-NN functions in knn.py with an example. Use the atomradii.csv and testing.csv that DSMCER used in for the inclass demo that relates atomic radii to atomic class. Leverage Markdown for your demo. (1 pt)
3. Create unit tests and put them in test_knn.py. There should be at least one unit test per function, though many more are appropriate for a real implementation. Again, use the atomradii data for the unit tests. Paste the output of running nosetests below. (2 pts)

Results of unit tests

Ran 5 tests in 3.673s

OK