IRIS Linear Regression using NumPy

run.py

import numpy as np
from numpy import genfromtxt

iris = genfromtxt("IRIS.csv", delimiter=",", skip_header=True)

training_examples = iris[:, list(range(3))]
targets = iris[:, -2]
bias_column = np.ones((training_examples.shape[0], 1))

inputs = np.hstack((bias_column, training_examples))

weights = np.zeros((4,))

weights[2] += 0.1


# Give predictions for all value
def predict(inputs, weights):
    return (np.dot(inputs, weights)).flatten()


# Calculate the loss of predictions
def loss(predictions, targets):
    return np.sum((predictions - targets) ** 2)


def gradient(rate, inputs, predictions, targets):
    return rate * np.dot(inputs.transpose(), predictions - targets) / targets.size


for i in range(1000):
    prediction = predict(inputs, weights)
    current_loss = loss(predict(inputs, weights), targets)
    print(current_loss)
    delta = gradient(0.01, inputs, predict(inputs, weights), targets)
    weights -= delta.transpose()

checker = np.hstack((inputs, prediction[:, None], targets[:, None]))
np.set_printoptions(precision=6)
print(checker)

This is a quick hack of a linear regression, based on what I read in https://www.cs.toronto.edu/~rgrosse/courses/csc321_2018/readings/L02%20Linear%20Regression.pdf