ML Data Distribution

Real machine learning projects need much larger datasets than a handful of numbers - this chapter covers generating and visualizing large random datasets.

Why Data Distribution Matters

Machine learning models learn general patterns, which requires seeing many examples, not just a handful. Understanding how your data is distributed - whether it is spread evenly, clustered around a center, or skewed to one side - helps you choose the right model and catch problems, like insufficient data or bad sampling, before training even begins.

Generating a Large Random Dataset

NumPy's random module can generate thousands of data points instantly, which is useful for experimenting with statistics and distributions before you have a real-world dataset. numpy.random.uniform() draws values evenly across a range, so every value in that range is equally likely to appear.

Example

import numpy as np

np.random.seed(42)
data = np.random.uniform(0.0, 10.0, 10000)

print('Number of points:', len(data))
print('Mean:', np.mean(data))
print('Std deviation:', np.std(data))
print('Min:', data.min(), 'Max:', data.max())

The Normal (Gaussian) Distribution

Many real-world measurements, such as heights, measurement errors, and test scores, roughly follow a normal distribution: a symmetric bell-shaped curve where values cluster around the mean and become rarer the farther they are from it. numpy.random.normal() generates data following this pattern, controlled by a mean (loc) and standard deviation (scale).

Example

import numpy as np

np.random.seed(42)
heights_cm = np.random.normal(loc=170, scale=8, size=10000)

print('Mean height:', np.mean(heights_cm))
print('Std deviation:', np.std(heights_cm))
within_one_std = np.mean(np.abs(heights_cm - np.mean(heights_cm)) <= np.std(heights_cm)) * 100
print('Approx percent of values within one std:', within_one_std)

Visualizing Distributions with Histograms

A histogram divides a dataset into equal-width bins and counts how many values fall into each bin, turning thousands of raw numbers into a shape you can interpret at a glance. Matplotlib's plt.hist() is the standard way to draw one, and choosing the right number of bins matters: too few bins hide detail, too many make the shape noisy.

Example

import numpy as np
import matplotlib.pyplot as plt

np.random.seed(42)
exam_scores = np.random.normal(loc=75, scale=10, size=10000)

plt.hist(exam_scores, bins=50)
plt.title('Distribution of Exam Scores')
plt.xlabel('Score')
plt.ylabel('Frequency')
plt.show()
DistributionNumPy functionShape
Uniformnp.random.uniform(low, high, size)Flat - every value equally likely
Normalnp.random.normal(loc, scale, size)Bell-shaped, symmetric around the mean
Binomialnp.random.binomial(n, p, size)Discrete, depends on n trials and probability p
  • Set a random seed (np.random.seed(42)) so your random dataset is reproducible
  • Larger sample sizes make histograms look smoother and closer to the true distribution
  • Skewed distributions with a long tail on one side often need the median rather than mean
  • Always plot your data before training a model - many issues are visible in a histogram first
Note: Use np.random.seed() before generating random data whenever you want the exact same 'random' numbers every time you rerun the script - essential for reproducible experiments and debugging.
Note: 10,000 or more data points is a common rule of thumb for a histogram to clearly reveal the true shape of a distribution; smaller samples can look noisy or misleading purely due to random chance.

Exercise: ML Data Distribution

Why do ML tutorials often use large random datasets to demonstrate data distribution?

Keep Going

Machine Learning Interview QuestionsPractice real questions asked for Machine LearningMachine Learning Engineer SalarySee what Machine Learning Engineer roles pay in the marketAuto Apply for JobsApply to open roles related to Machine Learning and tweak your resume