讲解k-Nearest、辅导data samples、辅导Python，Java，c/c++程序语言 解析Java程序|讲解R语言编程
- 首页 >> OS编程 Machine Learning, 2018-19
Coursework Part I
Description
The first part of your coursework will be based on your implementation of k-Nearest
Neighbours and nested cross-validation. The coursework is heavily based on your lab work, so
firstly make sure you complete the previous labs (up to and including Lab 4).
For the coursework, please make sure to implement your own code and not use libraries. You
will need to present your own code that performs nested cross-validation and the k-nearest
neighbour algorithm, build confusion matrices, and estimate distances between data samples.
Plagiarism: please make sure that the material you submit has been created by you. Any
sources you use for code should be properly referenced. Your code will be checked for
plagiarism using appropriate software.
Deliverables
The deliverables for your coursework should include:
1. A jupyter notebook, including the code you are submitting for the assignment.
2. A report (preferably PDF) that contains a write-up on your implementation, and
answers for a set of free-form questions (more details below).
The tasks for your coursework are based on the IRIS dataset, as used in the labs. Download the
coursework notebook that will guide you to complete the tasks for the coursework.
Note: Your code should run without issues on lab machines!
Report
Your report should include the following points discussed below. I‘m not giving a specific page
length limit, but it shouldn’t be longer than a couple of pages.
1. Write-up of your implementation. Include any details on your implementation of crossvalidation
and k-NN, details on the distance functions you used, and notes on any other code
you developed for the coursework. Highlight important points in your coursework here.
2. Results. Your report should contain:
? Tables showing overall test accuracy and standard deviation over 5-folds for (i) clean
data1
, and (ii) noisy data, while also indicating parameter choice (number of neighbours,
distance function).
A total confusion matrix over the 5-fold cross-validation for each of the runs (i.e. one
matrix for clean data, and another for noisy data). Each matrix can be the sum of the
matrices for each fold. There should be rows and columns for each class.
1 See example in Table 1 - the standard deviation (symbolized using ± in the table) is obtained via the
numpy function std. The total error is the average error over the 5 folds.Clean data accuracy k distance
Fold1 80 … …
Fold2 78 … …
Fold3 75 … …
Fold4 80 … …
Fold5 97 … …
total 82 ± 0.6 (empty) (empty)
Table 1. Example results table for clean data.
Results Analysis. Any conclusions you have by observing the behaviour of the
algorithm on clean and noisy data. You should explicitly try to answer the following
questions, justifying your answers based on the results.
(a) Do the best parameters change per fold? Can you say that one parameter choice
is better regardless of the data used?
(b) Does the best parameter choice change depending on whether we use clean or
noisy data? (Answer for both distance function and number of neighbours.)
3. Questions. Your report should also answer the following questions:
(a) Exploratory Data Analysis. What do you observe by plotting the figure for data
without noise? What do you observe when you add Gaussian noise and plot again?
(b) Tie breaking. Assume that you have selected the number of neighbours to be an even
number, e.g., 2. For one of the neighbours, the suggested class is 1, and for the other
neighbour the suggested class is 2. How would you break the tie? Write example
pseudocode that does this.
(c) Improving performance on noisy data. The performance of k-NN on the noisy data
should be worse than on the clean data. Suggest at least one way of improving the
performance on the noisy data. Try to elaborate on your idea as much as possible,
including pseudocode where possible.
Coursework Part I
Description
The first part of your coursework will be based on your implementation of k-Nearest
Neighbours and nested cross-validation. The coursework is heavily based on your lab work, so
firstly make sure you complete the previous labs (up to and including Lab 4).
For the coursework, please make sure to implement your own code and not use libraries. You
will need to present your own code that performs nested cross-validation and the k-nearest
neighbour algorithm, build confusion matrices, and estimate distances between data samples.
Plagiarism: please make sure that the material you submit has been created by you. Any
sources you use for code should be properly referenced. Your code will be checked for
plagiarism using appropriate software.
Deliverables
The deliverables for your coursework should include:
1. A jupyter notebook, including the code you are submitting for the assignment.
2. A report (preferably PDF) that contains a write-up on your implementation, and
answers for a set of free-form questions (more details below).
The tasks for your coursework are based on the IRIS dataset, as used in the labs. Download the
coursework notebook that will guide you to complete the tasks for the coursework.
Note: Your code should run without issues on lab machines!
Report
Your report should include the following points discussed below. I‘m not giving a specific page
length limit, but it shouldn’t be longer than a couple of pages.
1. Write-up of your implementation. Include any details on your implementation of crossvalidation
and k-NN, details on the distance functions you used, and notes on any other code
you developed for the coursework. Highlight important points in your coursework here.
2. Results. Your report should contain:
? Tables showing overall test accuracy and standard deviation over 5-folds for (i) clean
data1
, and (ii) noisy data, while also indicating parameter choice (number of neighbours,
distance function).
A total confusion matrix over the 5-fold cross-validation for each of the runs (i.e. one
matrix for clean data, and another for noisy data). Each matrix can be the sum of the
matrices for each fold. There should be rows and columns for each class.
1 See example in Table 1 - the standard deviation (symbolized using ± in the table) is obtained via the
numpy function std. The total error is the average error over the 5 folds.Clean data accuracy k distance
Fold1 80 … …
Fold2 78 … …
Fold3 75 … …
Fold4 80 … …
Fold5 97 … …
total 82 ± 0.6 (empty) (empty)
Table 1. Example results table for clean data.
Results Analysis. Any conclusions you have by observing the behaviour of the
algorithm on clean and noisy data. You should explicitly try to answer the following
questions, justifying your answers based on the results.
(a) Do the best parameters change per fold? Can you say that one parameter choice
is better regardless of the data used?
(b) Does the best parameter choice change depending on whether we use clean or
noisy data? (Answer for both distance function and number of neighbours.)
3. Questions. Your report should also answer the following questions:
(a) Exploratory Data Analysis. What do you observe by plotting the figure for data
without noise? What do you observe when you add Gaussian noise and plot again?
(b) Tie breaking. Assume that you have selected the number of neighbours to be an even
number, e.g., 2. For one of the neighbours, the suggested class is 1, and for the other
neighbour the suggested class is 2. How would you break the tie? Write example
pseudocode that does this.
(c) Improving performance on noisy data. The performance of k-NN on the noisy data
should be worse than on the clean data. Suggest at least one way of improving the
performance on the noisy data. Try to elaborate on your idea as much as possible,
including pseudocode where possible.