ECM3420/ECMM445

Posted 2021-03-24 codingjava

UNIVERSITY OF EXETER
COLLEGE OF ENGINEERING, MATHEMATICS, AND PHYSICAL SCIENCES
ECM3420/ECMM445
Learning from Data
Continuous Assessment
Date set: 14th November 2019
Hand-in date: 4th December 2019
This CA comprises 20% of the overall module assessment for the level 3 students and 10 % for the
M-Level students .
This is an individual exercise, and your attention is drawn to the guidelines on collaboration and
plagiarism in the College handbook.
See ELE for detailed submission instructions. This assignment requires you to make an electronic
submission using the Electronic Coursework Submission System (empslocal.ex.ac.uk/
cgi-bin/submit/prepare). The electronic submission should consist of a single Jupyter Notebook
(.ipynb) file containing the program code you are asked to produce and their outputs;
The questions are overleaf.
What you should submit
• A Jupyter Notebook file (.ipynb) containing the code and outputs for all questions.
See next page
1
1 Preamble
In this coursework you will have to implement a basic clustering analysis framework from scratch
(i.e., without using the Scikit-learn implementations) including both the algorithm and the validation
functions.
1. For this coursework you should not use the sklearn or mlxtend packages;
2. You can use other libraries such as pandas, numpy, scipy, and matplotlib ;
3. This coursework comprises 6 (six) work pieces WP.
2 Work specification
You will have to implement the three algorithms below:
(WP1) [25 marks] The k-means algorithm with the Euclidean distance
(WP2) [15 marks] The Davies-Bouldin index
(WP3) [15 marks] Silhouette score
Additionally, using your implementations, you will have to perform the analyses described
below on the provided data files: iris.txt, wine.txt and cluster_validation_data.txt:
(WP4) [15 marks] Perform model selection for selecting the partition order k generating a plot like
the one shown in Figure 1 using your implementation of the Davies-Bouldin index and analyze
the results commenting on: (1) the best number of partitions k for all the three datasets
and (2) For the wine and iris data, how well did your clustering algorithm performed compared
to the ground-truth known classes?
2 2.5 3 3.5 4 4.5 5 5.5 6
Number of clusters (k)
Davies-Bouldin values
Figure 1: Davies-Bouldin index
2
(WP5) [10 marks] Modify the function plot_silhouette provided in the lfd_utils.py file to use
your kmeans and silhouette_scores functions and generate a plot similar to Figure 2 for
each value of k (from 2 to max_k)
0.1 0.0 0.2 0.4 0.6 0.8 1.0
The silhouette coefficient values
Cluster label
0
1
The silhouette plot for the various clusters.
8 6 4 2 0 2
Feature space for the 1st feature
Feature space for the 2nd feature
The visualization of the clustered data.
Silhouette analysis for KMeans clustering on sample data with n_clusters = 2 Figure 2: Example of a Silhouette plot for k = 2
(WP6) [20 marks] Perform model selection for selecting the partition order k using the silhouette
plots for k from 2 to 6 commenting on: (1) how to interpret the silhouette scores for each
value of k taking into account the width of the partitions and how they compare with the
average silhouette score, (2) what would be the best number of partitions k for all the
three datasets and (3) For the wine and iris data, for the best k you found, how well did
your clustering algorithm performed compared to the ground-truth known classes?
3 Functions definitions
Your implementations for the k-means and silhouette score functions should have the following
signatures
• kmeans(x,k,max_itr=100)
– Parameters:
*
x: the data do be clustered
*
k: the number of clusters
*
max_itr: the maximum number of iterations
– Returns:
*
cluster_labels: the cluster membership labels for each element in the data x
• silhouette_scores(x, cluster_labels)
– Parameters:
*
x: the data
*
cluster_labels: the cluster membership vector produced by the k-means algorithm.
– Returns:
*
scores: a vector containing the silhouette score for each data sample in x.
3

因为专业，所以值得信赖。如有需要，请加QQ：99515681 或 微信：codehelp