slope_life_span.py

import numpy as np


def hypothesis(weights, x):
    return np.dot(weights, x)


def cost_derivative(weights, x, y, feature_x):
    s = 0
    m = len(x[0])
    for i in range(m):
        s += (hypothesis(weights, x[:, i]) - y[i]) * feature_x[i]
    return s / m


def train(x, y, a, number_of_iterations):
    i = 0
    n = len(x)  # number of features
    weights = [0] * n

    while i < number_of_iterations:
        temp_weights = [0] * n
        for j in range(n):
            temp_weights[j] = weights[j] - a * cost_derivative(weights, x, y, x[j])

        if temp_weights == weights:
            return weights

        for j in range(n):
            weights[j] = temp_weights[j]
        i += 1

    return weights


# slope degrees: for ex. 27° is converted to 2.7 to avoid overflow
slopes = [2.7, 2.1, 3.8, 1.8, 7.0, 6.8, 3.6, 4.9, 2.7, 5.0,
          4.7, 6.2, 4.4, 5.7, 5.5, 3.6, 6.9, 6.5, 6.7, 3.0,
          6.3, 4.1, 3.3, 2.3, 2.2, 4.8, 5.2, 3.6, 5.2, 3.3,
          1.8, 2.9, 4.3, 6.2, 3.5, 3.4, 2.7, 4.3, 5.4, 2.1]
# life spans: years with good condition until the first maintenance
output = [7.4, 9.5, 5.2, 11.1, 2.8, 2.9, 5.5, 4.0, 7.4, 4.0,
          4.2, 3.2, 4.5, 3.5, 3.6, 5.5, 2.8, 3.0, 2.9, 6.6,
          3.1, 4.8, 6.0, 8.6, 9.0, 4.1, 3.8, 5.5, 3.8, 6.0,
          11.1, 6.8, 4.6, 3.2, 5.7, 5.8, 7.4, 4.6, 3.7, 9.5]

bias = []
for i in range(40):
    bias.append(1)

input = np.array([bias, slopes])

w = train(input, output, 0.03, 1500)
print("weights = " + str(w)) # 11.178870048530353, -1.3534671476175058
years = hypothesis(w, [1, 2.2])
print("Estimate a slope life span with 22°: " + str(years)) # ~8.2 years