Homework 5

Due: Sunday, May 10, 2020, 11:55pm

Guidelines:

All homeworks will be submitted via Courseworks. For a given assignment, all files should be put in a folder called uni-hw5, all lowercase, which is then zipped and submitted on courseworks. The folder should contain a single readme.txt which has your name, uni, the homework number, and a brief summary of your work for each part (if necessary).

Nearest Neighbor

In this assignment we will complete our machine learning library. In addition to the functionality we have already written, we will implement the following features:

- Provide some statistics on the dataset using pandas:

- Skewness and kurtosis of each column in the dataset

- Mean/Std Deviation of each column, grouped into benign/malignant

- Plot some more advanced information using seaboard - Pairplot of each column against other columns

- Heatmap of the column correlations

- Implement the K nearest neighbors algorithm from class - Use scikit-learn to implement:

- K Nearest neighbors (for comparison) - Support Vector Machine classifier

Here is an example output from running %run -m engi1006. Note that the “...” indicate that I have omitted some output (e.g. for other columns).

   Please provide a filename: wdbc.csv

What percentage of the dataset would you like to put aside for testing? 20 What model would you like to run? (knn, sklearn-knn, svm) knn

Reading filename: wdbc.csv

Dataset size: 569 x 30

Number of benign: 357

Number of malignant: 212

Would you like to see more advanced stats? (y/n)y Column 0 statistics:

Skewness:0.9423795716731008 Kurtosis:0.8455216229065408 ...

Column 29 statistics:

Skewness:1.6625792663955172 Kurtosis:5.2446105558150125

    Dataframe statistics Benign Stats: Mean:

0

1

...

29

Name: B, dtype: float64 Std:

12.146524 17.914762

0.079442

0

1

...

29

Name: B, dtype: float64

Malignant Stats: Mean:

1.780512 3.995125

0.013804

0

1

...

29

Name: M, dtype: float64 Std:

17.462830 21.604906

0.091530

0

1

...

29

Name: M, dtype: float64

Would you like to plot the data? (y/n) y

How many columns should we plot in the scatter matrix? (5) 5

3.203971 3.779470

0.021553

     Splitting dataset into 80% for training and 20% for testing Test dataset has 113 entries

Train dataset has 456 entries

Hit enter to run algorithm

How many nearest neighbors? 5 Running knn classifier... Accuracy: 90.3%

Run again? (y/n) n

 Details

All functions have details about their implementation in their docstrings.