MultiDimensionalKNN

A k-nearest neighbours classifier that can operate on multidimensional data. It is created from scratch so no external modules are needed.

How to use the classifier

You can find the following code in the example.py file. To use the classifier first import it from the multiDKNN.py file:

from multiDKNN import MultiDKNN

Then create an instance of the classifier and fit it to your data:

clf = MultiDKNN() 
clf.fit(x_train,y_train)

Here x_train must be a list-like object that contains list-like objects with the features of your data points (e.g x_train = [[feature1,feature2,...],[feature1,feature2,...],...]). While y_train must be a list-like objects that contains your labels (e.g y_train = [label1,label2,...]). After that predict the labels of unlabeled data using:

outcomes = clf.predict(x_test,k)

Here x_test must be a list-like object that either contains other list-like objects (like x_train) if you want to predict the labels of multiple data points, or just the coordinates of one data points (e.g test_x = [13.4,45.7,231.23,89.0]). K represents the number of nearest data points that will be considered in order to determine the label(s) of the new data-point(s). If you want to calculate the accuracy of the classifier you can use:

accuracy =  clf.score(outcomes,y_test) 

where y_test must be a list-like object that contains the labels of the data used to predict the outcomes. In case you are not satisfied by the accuracy you can do:

best_k = clf.optimizek(x_test,y_test,up_limit) 

this will find the optimal value of K. Here test_x and test_y are as previously defined and up_limit is the highest value of K that will be tested. This method actually brute-forces all values of K up to up_limit so it might take a while.

How to use the included dataset

The multiDKNN.py file includes Fisher's(1936) iris dataset, formatted for imidiate use from the MultiDimensionalKNN classifier. The following code can be found in the example.py file. First of all import the dataset and create an instance of it:

from multiDKNN import IrisDataset()
data = IrisDataset()

Then you can seperate training and testing data with the following:

x_train = data.features[:140] 
y_train = data.labels[:140]  
x_test = data.features[140:] 
y_test = data.labels[140:]

This will results in the creation of four lists:

• x_train: A list of lists containig the coordinates of the first 140 datapoints.
• y_train: A list containing the labels of the first 140 datapoints.
• x_test: A list of lists containing the coordinates of the last 10 datapoints.
• y_test: A list containing the labels of the last 10 datapoints.

Of course you can change 140 to any number you like or use a more sophisticated method to get shuffled testing data. In case you want to see what each number in the features represents just run:

data.feature_names

References

FISHER, R. A. (1936), THE USE OF MULTIPLE MEASUREMENTS IN TAXONOMIC PROBLEMS. Annals of Eugenics, 7: 179-188. doi:10.1111/j.1469-1809.1936.tb02137.x