linear_regression_test.py

import linear_regression
import math_util
import random

def LinearRegressionTest(verbose=True):
    expected_weights = [200.0, 10.0, 200.0, -43.0, 1.0, -314.0]
    noise_stdev = 5.0
    training_data = CreateTrainingData(
        expected_weights, 1000, -100.0, 100.0, noise_stdev=noise_stdev)
    test_data = CreateTrainingData(
        expected_weights, 500, -100.0, 100.0, noise_stdev=noise_stdev)

    model = linear_regression.LinearRegression(len(expected_weights) - 1)
    model.RandomizeWeights(random_range=(-1000.0, 1000.0))
    model.Train(training_data,
                learning_rate=0.004,
                learning_iterations=10000,
                regularization_rate=0.0,
                verbose=verbose)

    total_difference = 0
    for t in test_data:
        total_difference += abs(model.Infer(t[0]) - t[1])
    print "learned model: %s" % (model,)
    print "average absolute difference on test data: %.3f" % (
        total_difference / len(test_data),)
    print "noise standard deviation around linear data: %.3f" % (noise_stdev,)
    
def CreateTrainingData(weights, num_training_examples, low, high,
                       noise_stdev=5.0):
    training_data = []
    for i in range(num_training_examples):
        input_data = [-1.0] + RandomVector(len(weights) - 1, low, high)
        expected_output = (
            math_util.VectorDotProduct(input_data, weights) +
            random.gauss(0.0, noise_stdev))
        training_data.append([input_data[1:], expected_output])
    return training_data
        
def RandomVector(size, low, high):
    return [random.uniform(low, high) for _ in range(size)]