Day 6: Statistics Fundamentals for AI Engineers

Welcome to Day 6 of Math Week. It is not enough to simply train a neural network; you must prove your model actually works and didn't just get lucky. To do this, we use Statistics.

Today, we cover Measures of Central Tendency (Mean, Median, Mode), Measures of Dispersion (Variance, Standard Deviation), and Hypothesis Testing.

Central Tendency & Dispersion

You already know what an average is. But in Data Science, we need to know how "spread out" our data is. If the average exam score is 50%, did everyone get exactly 50%? Or did half the class get 0% and half get 100%?

Variance and Standard Deviation measure this spread.

We can calculate all of this instantly using NumPy!

# day_ex1.py
import numpy as np

# A small dataset of exam scores
data = [10, 20, 30, 40, 50]

# Central Tendency
mean = np.mean(data)

# Dispersion
variance = np.var(data)
std_dev = np.std(data)

print("Mean: ", mean)
print("Variance: ", variance)
print("Standard Deviation: ", std_dev)

Hypothesis Testing & P-Values

Imagine you train an AI model. Model A predicts housing prices with an error of $1000. Model B predicts them with an error of $900. Is Model B actually better? Or could that tiny difference just be random luck on the testing set?

We prove it using Hypothesis Testing.

There are two hypotheses: * Null Hypothesis (H0): There is NO difference between the models. Any difference is pure luck. * Alternative Hypothesis (H1): Model B is legitimately, statistically better.

When we run a statistical "T-Test," it spits out a number called a P-Value, which represents the probability that the Null Hypothesis is true. In Science and ML, we usually say that if the P-Value is < 0.05 (meaning there is less than a 5% chance the results were luck), we "Reject" the null hypothesis, and accept that our AI is actually better!

Let's look at day6_ex2.py and run a T-Test on two groups using the scipy.stats library.

# day6_ex2.py
from scipy.stats import ttest_ind

# Did Group 2 score significantly higher than Group 1?
group1 = [2.1, 2.5, 2.8, 3.0, 3.2]
group2 = [1.8, 2.0, 2.4, 2.7, 2.9]

# Perform independent t-test
t_stat, p_value = ttest_ind(group1, group2)

print("T-Statistic: ", t_stat)
print("P-Value: ", p_value)

# Interpretation!
alpha = 0.05
if p_value < alpha:
    print("Reject the null hypothesis: There is a significant difference!")
else:
    print("Failed to reject the null hypothesis: It might just be random luck!")

Wrapping Up Day 6

You are now equipped with the statistical tools to rigorously evaluate data and machine learning models. You can calculate confidence intervals, standard deviations, and prove your findings with P-Values.

Tomorrow is the big one. Day 7: Linear Regression from Scratch.

We are going to combine the Linear Algebra matrices from Day 1, the Calculus gradients from Day 3, and the Statistical Mean-Squared-Error from Day 6 into a single project! See you then!