minte9

LearnRemember

Diabetes

 
""" Diabetes dataset / Knn Algorithm

Target is a quantitative measure of disease progression 
one year after baseline. The features are: 
age, sex, body mass index, average blood, etc.

The target values are continuous not categorical.
We need to divide the target values into two categories.

1 represents target value greater than the mean target value
0 represents target value less than or equal
"""

import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_diabetes

# Diabetes dataset
dataset = load_diabetes()
X = dataset.data
y = dataset.target
features = dataset.feature_names

X = X[:, [2, 3]]                        # select 2 features to plot (bmi, bp)
y_binary = np.where(y > y.mean(), 1, 0) # make the target categorical
class_names = ['OK', 'NOK']             # 0 is 'OK', 1 is 'NOK'


# Algorithm
def knn(x_unknown, k=5):
    distances = np.sqrt(np.sum((X - x_unknown)**2, axis=1))

    keys = np.argsort(distances)
    knn_keys = keys[:k]

    knn_classes = y_binary[knn_keys]
    knn_classes = knn_classes.astype(int)

    knn_most_common_class = np.bincount(knn_classes).argmax()
    knn_class = class_names[knn_most_common_class]
    return knn_class, knn_keys

# Unknown points
xA = np.array([-0.075, 0.040])
xB = np.array([0.10, 0.035])
k_nearest = 5 # number of nearest neighbors (k=5)

# Predictions
knn_classA, knn_keysA = knn(xA, k_nearest)
knn_classB, knn_keysB = knn(xB, k_nearest)


# Plot the two unknown points and dataset points
fig, ax = plt.subplots()
ax.set_xlabel('bmi (body mass index)')
ax.set_ylabel('bp (average blood pressure)') 

plt.scatter(X[:, 0], X[:, 1], c=y_binary)
plt.scatter(xA[0], xA[1], marker='o', color='g', label=knn_classA)
plt.scatter(xB[0], xB[1], marker='x', color='r', label=knn_classB)

# Plot lines to the nearest points
for i in knn_keysA:
    plt.plot((xA[0], X[i][0]), (xA[1], X[i][1]), color='gray', linestyle='--')
for i in knn_keysB:
    plt.plot((xB[0], X[i][0]), (xB[1], X[i][1]), color='gray', linestyle='--')

plt.title('Diabetes dataset - Knn')
plt.legend()
plt.show()

print("Keys:", dataset.keys())
print("Shape:", dataset['data'].shape)
print("Target values examples:", dataset.target[0:7])
print(dataset['DESCR'])

"""
    Keys: dict_keys(['data', 'target', 'frame', 'DESCR', ...])
    Shape: (442, 10)
    Target values examples: [151.  75. 141. 206. 135.  97. 138.]
    Diabetes dataset
    ----------------
    Ten baseline variables, age, sex, body mass index, average blood
    pressure, and six blood serum measurements were obtained for each of n =
    442 diabetes patients, as well as the response of interest, a
    quantitative measure of disease progression one year after baseline.
"""