Knn classification on a dataset

Question

Mary Gh 2020-12-31

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/706058-knn-classification-on-a-dataset

评论： merlin toche 2023-2-16

toy_data_csv.zip

Hi i have this dataset and i want knn classification for this and also find accuracy of performance of this classification and showing the number of wrong and true classifications with confusion matrix. Any one can help me with this .please.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Mary Gh 2020-12-31

Well they are A to I classes with 2 train datas (two lables,)and one test data

Mary Gh 2020-12-31

Actually i dont think they represent anything it didnt say in question

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Image Analyst 2020-12-31

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/706058-knn-classification-on-a-dataset#answer_588548

kNearestNeighbor.m

So one is to train, the other "training" one is for validation (run it through and see how accurate the predictions are compared to the known values), and the third one is a test set (which you do not know the correct answers for).

I have attached a KNN demo. See if you can adapt that to your homework problem. If not, come back for more hints.

21 个评论
显示 19更早的评论隐藏 19更早的评论

Image Analyst 2022-12-3

在 MATLAB Online 中打开

I don't know what this means

x_Train = [4 6 7 5 8; 5.2 6.3 9 11 10]
x_Train = 2×5
    4.0000    6.0000    7.0000    5.0000    8.0000
    5.2000    6.3000    9.0000   11.0000   10.0000
y_Train = [3 7 8 5 8;4.5 1.3 6 7 9.1]
y_Train = 2×5
    3.0000    7.0000    8.0000    5.0000    8.0000
    4.5000    1.3000    6.0000    7.0000    9.1000
x_test=[0.8 14 2 5 4.3; 7.2 6.5 4.1 18 3.6]
x_test = 2×5
    0.8000   14.0000    2.0000    5.0000    4.3000
    7.2000    6.5000    4.1000   18.0000    3.6000
y_test=[1 4.8 5.9 14 3.4 ;9 17 12 16 2.9]
y_test = 2×5
    1.0000    4.8000    5.9000   14.0000    3.4000
    9.0000   17.0000   12.0000   16.0000    2.9000
trainclass=['None','OCF','SCF','P','SBD','OCI']
trainclass = 'NoneOCFSCFPSBDOCI'

Why are there two rows in your x and y values?

Do you think you're getting kmeans confused with KNN?

The way KNN works is you specify K, the data, and the class number that each data point belongs to (it's "TRUE" class assignment). For example if you had 100 data training points (100 x and 100 y in vectors) and 5 classes, then you must have a training vector that says what class number each of those 100 training points really is. For example if trainingClasses = [1,3,5,2,.......3,2] then it says training point 1 is defined to be a member of class 1, and training point 2 is defined to be a member of class 3, and training point 3 is defined to be a member of class 5, and training point 4 is defined to be a member of class 2, and ... training point 99 is defined to be a member of class 3, and training point 100 is defined to be a member of class 2.

Now you can call knnsearch and it will tell you the K indexes in your training set that each point in your test set is closest to

%-----------------------------------------------------------------------------------------------------------------
% Now do a K Nearest Neighbor Search.
% Get the classes of the unknown data.
% First collect all the training data into one tall array
trainingCoords = [x_Train(:), y_Train(:)]
trainingCoords = 10×2
0000    3.0000
2000    4.5000
0000    7.0000
3000    1.3000
0000    8.0000
0000    6.0000
0000    5.0000
0000    7.0000
0000    8.0000
0000    9.1000
unknownCoords = [x_test(:), y_test(:)]
unknownCoords = 10×2
8000    1.0000
2000    9.0000
0000    4.8000
5000   17.0000
0000    5.9000
1000   12.0000
0000   14.0000
0000   16.0000
3000    3.4000
6000    2.9000
[indexes, distancesOfTheIndexes] = knnsearch(trainingCoords, unknownCoords, ...
    'NSMethod', 'exhaustive',...
    'k', 5,... % Get indexes of the 5 nearest points
    'distance', 'euclidean') % Regular Pythagorean formula for distance
indexes = 10×5
   4     2     7     3
   9     3    10     6
   6    10     9     5
   5     9     3     8
   2     1     3     5
   3     9    10     7
   9    10     3     6
   8     9     6     5
   2     7     4     3
   2     7     4     3
distancesOfTheIndexes = 10×5
7736    5.5082    5.6223    5.8000    7.9398
0198    1.2806    2.3324    2.8018    3.4986
7202    5.1420    5.8728    6.8000    7.6968
6406    9.0139    9.1241   10.0125   10.9659
1321    3.4928    3.5228    4.1485    5.4231
9406    5.3488    5.5866    6.5742    7.0576
3246    6.7082    7.0007    7.0711    8.9443
5646   11.4018   12.8062   13.4536   13.6015
5000    1.4213    1.7464    2.9000    3.9812
4123    2.2627    2.5239    3.1385    4.7508
% Extract the classes

Now you can see that point 1 of the test set is closest to points 1, 4, 2, 7, and 3 of the training set, in order of decreasing closeness (increasing distance). If I wanted to determine the classes of those 5 training points, I need to know what class they are in -- their "true" class that they are known to be without any doubt. For example if points 1, 4, and 2 are all in class 3, then the majority of points near test point 1 are class 3 so we're now going to define test point 1 as being in class 3 also.

That's KNN. Does that explain it?

Now with kmeans, you don't have a training set because it's an upsupervised classification. All you have is a training set and a known/desired number of clusters to force them into. So it will try to figure out where the clusters are and assign the unknown points into one of the k clusters. But there is not ground truth, or training set, where you know the true class for any points.

merlin toche 2023-1-6

thank you sir! i'm very happy.

please, sir I have other questions via my profile as you said concern your previous answers.

best regards

Image Analyst 2023-1-6

@merlin toche Not sure what that means, but in your profile, it says you've submitted only one question, and one answer. If you have asked other questions in the comments to someone else's question, then that won't show up. That's why it's best to ask your own questions in your own thread rather than try to post the questions in someone else's thread, like here in @Mary Gh's thread. She'll get emails every time there is activity in this thread.

请先登录，再进行评论。

Answer 2

merlin toche 2022-12-9

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/706058-knn-classification-on-a-dataset#answer_1123957

thank you sir for all you do for me. well explained and well understood. excuse me for continuing to bother you, please, I'm still learning machine learning, and you are a good teacher for me. I have two concerns sir:

My first question is sir after partitioning my data (I did it using cvpartion mydata1=rand(150.4);

% cross-variation (Train: 80%, Test: 20%)

cv=cvpartition(size(mydata1,1),'holdout',0.2);

idx=cv.test;

% sparse training and test data

mydata1_train=mydata1(~idx,:);

mydata1_test=mydata1(idx,:);

% data evaluation), which command used to do the job you just explained to me? for example with my 150 data partitioned, would I still need to declare the vectors below before doing the work you explained to me? if not how i have to call this data in order to build it?

x_Train = [4 6 7 5 8; 5.2 6.3 9 11 10];

y_Train = [3 7 8 5 8;4.5 1.3 6 7 9.1];

x_test=[0.8 14 2 5 4.3; 7.2 6.5 4.1 18 3.6];

y_test=[1 4.8 5.9 14 3.4 9 17 12 16 2.9];

['None', 'OCF', 'SCF', 'P', 'SBD', 'OCI']

best regards

4 个评论
显示 2更早的评论隐藏 2更早的评论

Image Analyst 2023-1-4

在 MATLAB Online 中打开

You have this:

x_train = mydata1_train(:,1); %[4 6 7 5 8];
y_train = mydata1_train(:,2); % [3 7 8 5 8];
% Now you say your classes are c=['open,'short','short','open','open']
% so let's make those class numbers.
trainClass =[1,2,2,1,1];

However your x_train and y_train have 120 elements. So you need to define trainClass to have 120 elements also. You need to know the "true" class for every one of your training points. Right now you have only 5, not 120.

merlin toche 2023-2-16

hi @Image Analyst

please can someone help me?

I want to detect a series of faults using the fuzzy-KNN algorithm. for this I have 5 name data classes, I wrote a code but errors appear, I would like your help to reread and make the necessary corrections.

attached my code and my dataset

THANKS

请先登录，再进行评论。

Knn classification on a dataset

3 个评论
显示 1更早的评论隐藏 1更早的评论

采纳的回答

21 个评论
显示 19更早的评论隐藏 19更早的评论

更多回答（1 个）

4 个评论
显示 2更早的评论隐藏 2更早的评论

另请参阅

类别

标签

产品

Community Treasure Hunt

Knn classification on a dataset

3 个评论 显示 1更早的评论隐藏 1更早的评论

采纳的回答

21 个评论 显示 19更早的评论隐藏 19更早的评论

更多回答（1 个）

4 个评论 显示 2更早的评论隐藏 2更早的评论

另请参阅

类别

标签

产品

Community Treasure Hunt

3 个评论
显示 1更早的评论隐藏 1更早的评论

21 个评论
显示 19更早的评论隐藏 19更早的评论

4 个评论
显示 2更早的评论隐藏 2更早的评论