Example usage

To use pkg_pyknnclassifier in a project:

import pkg_pyknnclassifier
import pandas as pd

from pkg_pyknnclassifier.data_loading import data_loading
from pkg_pyknnclassifier.scaling import scaling
from pkg_pyknnclassifier.predict import predict
from pkg_pyknnclassifier.evaluate import evaluate

Check the version of our package.

print(pkg_pyknnclassifier.__version__)

0.1.0

Narrative

We will now conduct an initial data wrangling to make our data ready for classification. Note our package was designed to only accept numerical features, and the binary target variable. If your data involves some categorical features, please convert them into a numerical format using appropriate encoding techniques before using our package.

Data-Loading

For better interpretation, we used Iris dataset, which comprises 2 distinct categories (Iris Setosa and Iris Versicolor), each with 50 samples, along with their corresponding features: “SepalLengthCm”: length of the sepal in cm “SepalWidthCm”: width of the sepal in cm “PetalLengthCm”: length of the petal in cm “PetalWidthCm” width of the petal in cm

The data showcases various species of the iris flower, serving as a prime example to demonstrate and evaluate the performance of our tailored k-Nearest Neighbors (kNN) model.

We load the dataset using the data_loading() function, importing the ‘Iris.csv’ file and specifying the target column upon which the categorization is based.

path_to_training = "../data/iris_train.csv"
path_to_validation = "../data/iris_valid.csv"
target_column = "Species"
X_train, y_train = data_loading(path_to_training, target_column)

To give a glance at our dataset:

X_train.head()

	SepalLengthCm	SepalWidthCm	PetalLengthCm	PetalWidthCm
0	4.8	3.4	1.6	0.2
1	4.6	3.2	1.4	0.2
2	7.7	3.0	6.1	2.3
3	5.2	3.4	1.4	0.2
4	6.8	3.2	5.9	2.3

We have excluded the “id” column, recognizing it as a unique identifier that could lead to overfitting if included. Since it does not offer any meaningful predictive value, it is appropriately removed from the feature set for our analysis.

X_val, y_val = data_loading(path_to_validation, target_column)
print(set(list(y_train)))

{'Iris-setosa', 'Iris-virginica'}

Scaling

Before building our kNN model, we scale our features so no features dominate the distance calculations. The model can learn from all the features equally and leads to a better performing model. Here we used the ‘MinMaxScaler’. ‘StandardScaler’ can be an alternative by filling it into scale_method parameter.

In addition, in this dataset, we do not have any missing value, but you can specify the impute strategy if there is any missing value in your dataset.

train_X_scaled = scaling(X_train, impute_strategy="mean", scale_method="MinMaxScaler")

Prediction

The line below is to predict the labels of the unlabeled observations based on the similarity score from the Euclidean distance.

y_pred = predict(X_train, y_train, X_val, pred_method="hard", k=3)
pd.DataFrame({'predictions': y_pred, 'values': y_val})

	predictions	values
0	Iris-virginica	Iris-virginica
1	Iris-virginica	Iris-virginica
2	Iris-virginica	Iris-virginica
3	Iris-setosa	Iris-setosa
4	Iris-setosa	Iris-setosa
5	Iris-setosa	Iris-setosa
6	Iris-setosa	Iris-setosa
7	Iris-virginica	Iris-virginica
8	Iris-setosa	Iris-setosa
9	Iris-setosa	Iris-setosa
10	Iris-setosa	Iris-setosa
11	Iris-setosa	Iris-setosa
12	Iris-virginica	Iris-virginica
13	Iris-setosa	Iris-setosa
14	Iris-virginica	Iris-virginica
15	Iris-setosa	Iris-setosa
16	Iris-virginica	Iris-virginica
17	Iris-virginica	Iris-virginica
18	Iris-setosa	Iris-setosa
19	Iris-setosa	Iris-setosa
20	Iris-virginica	Iris-virginica
21	Iris-virginica	Iris-virginica
22	Iris-setosa	Iris-setosa
23	Iris-setosa	Iris-setosa
24	Iris-virginica	Iris-virginica
25	Iris-setosa	Iris-setosa
26	Iris-setosa	Iris-setosa
27	Iris-virginica	Iris-virginica
28	Iris-setosa	Iris-setosa
29	Iris-virginica	Iris-virginica

Evaluate

We try to evaluate how our model is performing, and by this line of code, we can see the accuracy of how our model fits to the validation dataset, and it seems that this model fits really well to this dataset. Alternatively, other metrics (such as precision, recall, and f1 score) can be used to evaluate the performance.

y_pred_train = predict(X_train, y_train, X_train, pred_method="hard", k=3)
acc_train = evaluate(y_train, y_pred_train, metric='accuracy')
print("The training score is:", acc_train)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[8], line 2
      1 y_pred_train = predict(X_train, y_train, X_train, pred_method="hard", k=3)
----> 2 acc_train = evaluate(y_train, y_pred_train, metric='accuracy')
      3 print("The training score is:", acc_train)

File ~/checkouts/readthedocs.org/user_builds/pkg-pyknnclassifier/checkouts/latest/src/pkg_pyknnclassifier/evaluate.py:35, in evaluate(y_true, y_pred, metric)
     31 # Check if y_true and y_pred are of correct types
     32 if not isinstance(y_true, (list, np.ndarray)) or not isinstance(
     33     y_pred, (list, np.ndarray)
     34 ):
---> 35     raise TypeError("y_true and y_pred must be lists or numpy arrays.")
     37 # Check if predicted and true labels have the same length
     38 if len(y_true) != len(y_pred):

TypeError: y_true and y_pred must be lists or numpy arrays.

acc_valid = evaluate(y_val, y_pred, metric='accuracy')
print("The validation score is:", acc_valid)

The validation score is: 1.0

Reference

Fisher,R. A.. (1988). Iris. UCI Machine Learning Repository. https://doi.org/10.24432/C56C76.