Probability distributions in MATLAB - Deep Dive

Overview - Probability distributions

What is it?

Probability distributions describe how likely different outcomes are in a random process. They assign a probability to each possible result, showing which outcomes are common and which are rare. This helps us understand uncertainty and make predictions based on data. Distributions can be shown as formulas, tables, or graphs.

Why it matters

Without probability distributions, we would not know how to measure uncertainty or predict future events based on past data. They help in fields like weather forecasting, quality control, and risk management. Without them, decisions would be guesses without a clear sense of risk or chance.

Where it fits

Before learning probability distributions, you should understand basic probability concepts like events and outcomes. After this, you can learn statistical inference, hypothesis testing, and machine learning models that rely on these distributions.

Mental Model

Core Idea

A probability distribution is a map that tells you how likely each possible outcome is in a random situation.

Think of it like...

Imagine a spinner divided into colored sections of different sizes. The bigger a section, the more likely the spinner will land there. The probability distribution is like the spinner’s color map showing the chance of landing on each color.

┌─────────────────────────────┐
│      Probability Distribution│
├─────────────┬───────────────┤
│ Outcome     │ Probability    │
├─────────────┼───────────────┤
│ A           │ 0.5           │
│ B           │ 0.3           │
│ C           │ 0.2           │
└─────────────┴───────────────┘

Build-Up - 7 Steps

1

Foundation

Understanding Random Events

Concept: Learn what random events and outcomes are in probability.

A random event is something that can have different results each time you try it, like flipping a coin or rolling a die. Each possible result is called an outcome. For example, flipping a coin has two outcomes: heads or tails.

Result

You can identify random events and list their possible outcomes.

Understanding what random events and outcomes are is the base for all probability concepts.

2

Foundation

Defining Probability Basics

Concept: Learn how to assign probabilities to outcomes.

Probability is a number between 0 and 1 that shows how likely an outcome is. If something is impossible, its probability is 0. If it is certain, its probability is 1. For example, the probability of rolling a 3 on a fair six-sided die is 1/6.

Result

You can calculate simple probabilities for events.

Knowing how to assign probabilities helps quantify uncertainty in random events.

3

Intermediate

Discrete Probability Distributions

🤔Before reading on: do you think a discrete distribution can assign probabilities to any number, or only specific values? Commit to your answer.

Concept: Discrete distributions assign probabilities to countable outcomes.

Discrete distributions describe random variables that take specific values, like the number of heads in coin flips. The probabilities for all possible values add up to 1. Examples include the Binomial and Poisson distributions.

Result

You can describe and calculate probabilities for discrete random variables.

Understanding discrete distributions allows modeling of countable outcomes common in real life.

4

Intermediate

Continuous Probability Distributions

🤔Before reading on: do you think continuous distributions assign probabilities to exact values or ranges? Commit to your answer.

Concept: Continuous distributions describe variables that can take any value in a range.

Continuous distributions use a function called a probability density function (PDF) to describe probabilities over intervals. The total area under the PDF curve is 1. Examples include the Normal (bell curve) and Uniform distributions.

Result

You can understand and work with probabilities for continuous data.

Knowing continuous distributions is key for modeling measurements like height or temperature.

5

Intermediate

Common Probability Distributions

Concept: Learn about popular distributions and their uses.

Some common distributions are: - Binomial: counts successes in fixed trials - Poisson: counts rare events in time/space - Normal: models natural variations - Uniform: all outcomes equally likely Each has formulas and shapes that fit different real-world data.

Result

You can recognize and choose distributions for different problems.

Knowing common distributions helps match data to the right model for analysis.

6

Advanced

Using MATLAB for Distributions

🤔Before reading on: do you think MATLAB can generate random numbers from distributions or only calculate probabilities? Commit to your answer.

Concept: Learn how to use MATLAB functions to work with distributions.

MATLAB has built-in functions like binopdf, poisspdf, normpdf, and rand to calculate probabilities and generate random samples. For example, normpdf(x,mu,sigma) gives the Normal PDF value at x. randn generates Normal random numbers.

Result

You can calculate probabilities and simulate data using MATLAB.

Using MATLAB tools makes working with distributions practical and efficient.

7

Expert

Understanding Distribution Properties

🤔Before reading on: do you think mean and variance fully describe all distributions? Commit to your answer.

Concept: Explore deeper properties like moments, skewness, and kurtosis.

Distributions have properties beyond mean and variance, such as skewness (asymmetry) and kurtosis (tail heaviness). These affect shape and behavior. Experts use these to better fit models and understand data nuances.

Result

You can analyze and compare distributions beyond basic statistics.

Knowing advanced properties helps in fine-tuning models and detecting unusual data patterns.

Under the Hood

Probability distributions work by assigning a probability measure to subsets of possible outcomes. For discrete variables, this is a probability mass function (PMF) giving exact probabilities. For continuous variables, a probability density function (PDF) defines probabilities over intervals via integration. Internally, these functions satisfy rules like non-negativity and total probability summing to one, ensuring consistent probability assignments.

Why designed this way?

Distributions were designed to formalize uncertainty mathematically, allowing consistent reasoning about random events. Early mathematicians needed a way to model chance phenomena precisely. The split into discrete and continuous reflects the nature of data types encountered. Alternatives like frequency counts lacked generality and predictive power.

┌───────────────┐
│ Random Event  │
└──────┬────────┘
       │
       ▼
┌───────────────┐
│ Probability   │
│ Distribution  │
├───────────────┤
│ PMF (discrete)│
│ or PDF (cont) │
└──────┬────────┘
       │
       ▼
┌───────────────┐
│ Probabilities │
│ for outcomes  │
└───────────────┘

Myth Busters - 3 Common Misconceptions

Quick: Do you think the probability of a continuous variable taking an exact value can be greater than zero? Commit to yes or no.

Common Belief: A continuous random variable can have a positive probability at a single exact value.

Reality: For continuous variables, the probability of any exact value is zero; probabilities are only meaningful over intervals.

Why it matters: Believing otherwise leads to incorrect probability calculations and misunderstandings of continuous data behavior.

Quick: Do you think the sum of probabilities in a distribution can be more than 1? Commit to yes or no.

Common Belief: Probabilities for all outcomes can add up to more than 1 if events overlap.

Reality: The total probability for all possible outcomes must always equal exactly 1, ensuring a valid distribution.

Why it matters: Ignoring this leads to invalid models and wrong predictions.

Quick: Do you think the mean alone fully describes a distribution's shape? Commit to yes or no.

Common Belief: Knowing the mean is enough to understand a distribution completely.

Reality: Mean shows central tendency but does not capture spread, skewness, or tail behavior.

Why it matters: Relying only on mean can hide important data features and cause poor decisions.

Expert Zone

1

Some distributions look similar but differ in tail behavior, affecting risk assessment in finance or engineering.

2

Parameter estimation methods (like maximum likelihood) can bias results if assumptions about distributions are wrong.

3

Mixture models combine multiple distributions to model complex data, a technique often missed by beginners.

When NOT to use

Probability distributions assume data randomness and independence; they are not suitable for deterministic or highly dependent data. In such cases, use deterministic models or time series analysis instead.

Production Patterns

In production, distributions are used for anomaly detection by modeling normal behavior, in simulations to generate synthetic data, and in Bayesian inference to update beliefs with new data.

Connections

Statistics

Probability distributions provide the foundation for statistical inference and hypothesis testing.

Understanding distributions helps interpret statistical results and confidence intervals correctly.

Machine Learning

Many machine learning algorithms assume data follows certain probability distributions to make predictions.

Knowing distributions improves model selection and evaluation in machine learning.

Quantum Mechanics

Quantum mechanics uses probability distributions to describe the likelihood of particle states.

Recognizing probability distributions in physics shows their universal role in describing uncertainty.

Common Pitfalls

#1 Confusing probability mass function (PMF) with probability density function (PDF).

Wrong approach:

Using PDF values as probabilities for exact outcomes in continuous data, e.g., P(X=2) = pdf(2).

Correct approach:

Calculate probabilities over intervals using integration, e.g., P(1.9 < X < 2.1) = integral of pdf from 1.9 to 2.1.

Root cause:

Misunderstanding that PDFs represent densities, not direct probabilities.

#2 Forgetting to normalize probabilities so they sum to 1.

Wrong approach:

Assigning probabilities that add up to 1.2 or 0.8 in a distribution.

Correct approach:

Adjust probabilities so total sum equals exactly 1.

Root cause:

Ignoring the fundamental rule that total probability must be one.

#3 Assuming all data fits a Normal distribution.

Wrong approach:

Modeling skewed or discrete data with Normal distribution without checking fit.

Correct approach:

Choose distribution based on data characteristics, e.g., Binomial for counts, not Normal.

Root cause:

Overgeneralizing Normal distribution due to its popularity.

Key Takeaways

Probability distributions assign likelihoods to possible outcomes, helping us understand and predict uncertainty.

Discrete distributions handle countable outcomes, while continuous distributions describe ranges of values.

Common distributions like Binomial and Normal fit different real-world data types and problems.

MATLAB provides tools to calculate probabilities and simulate data from these distributions easily.

Advanced properties like skewness and kurtosis reveal deeper insights into data shape beyond mean and variance.

Practice

1. Which MATLAB function is used to create a probability distribution object?

easy

A. rand

B. makedist

C. histogram

D. plot

Solution

1. Step 1: Understand the purpose of each function

   makedist creates a probability distribution object, while rand generates random numbers, histogram plots data distribution, and plot creates graphs.

2. Step 2: Identify the correct function for creating distributions

   Only makedist is designed to create probability distribution objects in MATLAB.

3. Final Answer:

   makedist -> Option B

4. Quick Check:

   Function to create distribution = makedist [OK]

Hint: Use makedist to create distributions in MATLAB [OK]

Common Mistakes:

• Confusing makedist with rand
• Using histogram to create distributions
• Trying to use plot for distributions

2. Which of the following is the correct syntax to create a normal distribution with mean 0 and standard deviation 1 in MATLAB?

easy

A. pd = makedist('normal', 0, 1);

B. pd = makedist('Normal', 'mean', 0, 'std', 1);

C. pd = makedist('Normal', 0, 1);

D. pd = makedist('Normal', 'mu', 0, 'sigma', 1);

Solution

1. Step 1: Recall the correct parameter names for normal distribution

   In MATLAB, the normal distribution parameters are specified as 'mu' for mean and 'sigma' for standard deviation.

2. Step 2: Check the syntax for makedist

   The correct syntax uses name-value pairs: makedist('Normal', 'mu', 0, 'sigma', 1). Other options use incorrect parameter names or formats.

3. Final Answer:

   pd = makedist('Normal', 'mu', 0, 'sigma', 1); -> Option D

4. Quick Check:

   Normal dist params = 'mu' and 'sigma' [OK]

Hint: Use 'mu' and 'sigma' as parameter names for normal dist [OK]

Common Mistakes:

• Using lowercase 'normal' instead of 'Normal'
• Using 'mean' and 'std' instead of 'mu' and 'sigma'
• Passing parameters without name-value pairs

3. What is the output of the following MATLAB code?

pd = makedist('Exponential', 'mu', 2);
p = pdf(pd, 1);
disp(round(p, 3));

medium

A. 0.303

B. 0.368

C. 0.500

D. 0.607

Solution

1. Step 1: Understand the exponential distribution PDF formula

   The PDF for exponential with mean mu=2 is f(x) = (1/2)*exp(-x/2). For x=1, f(1) = 0.5 * exp(-0.5) ≈ 0.5 * 0.6065 = 0.3033.

2. Step 2: Compute and round the value

   p ≈ 0.3033, round(p, 3) = 0.303.

3. Final Answer:

   0.303 -> Option A

4. Quick Check:

   (1/mu)*exp(-x/mu) at x=1, mu=2 ≈ 0.303 [OK]

Hint: Recall pdf = (1/mu)*exp(-x/mu) for exponential [OK]

Common Mistakes:

• Ignoring the (1/mu) factor
• Confusing PDF with CDF
• Rounding errors causing mismatch

4. Identify the error in this MATLAB code snippet:

pd = makedist('Binomial', 'n', 10, 'p', 0.5);
prob = cdf(pd, 11);

medium

A. The probability p must be an integer

B. makedist does not support 'Binomial' distribution

C. The cdf input value 11 exceeds the number of trials n

D. The variable name 'prob' is reserved

Solution

1. Step 1: Understand binomial distribution parameters

   Binomial distribution with n=10 means possible outcomes are from 0 to 10.

2. Step 2: Check the cdf input value

   Calling cdf(pd, 11) asks for probability of outcomes ≤ 11, but 11 is outside the valid range (max 10).

3. Step 3: Identify the error

   Input to cdf must be within 0 to n. Using 11 is invalid and causes an error or unexpected result.

4. Final Answer:

   The cdf input value 11 exceeds the number of trials n -> Option C

5. Quick Check:

   cdf input must be ≤ n for binomial [OK]

Hint: cdf input must be within distribution support range [OK]

Common Mistakes:

• Using cdf input outside valid range
• Thinking p must be integer
• Assuming makedist lacks binomial support

5. You want to calculate the probability that a normally distributed variable with mean 5 and standard deviation 2 is less than 7 in MATLAB using a probability distribution object. Which code correctly computes this?

hard

A. pd = makedist('Normal', 'mu', 5, 'sigma', 2); p = cdf(pd, 7);

B. p = normcdf(7, 5, 2);

C. pd = makedist('Normal', 5, 2); p = pdf(pd, 7);

D. p = cdf('Normal', 7, 5, 2);

Solution

1. Step 1: Understand how to create a normal distribution object

   Use makedist('Normal', 'mu', 5, 'sigma', 2) to create the distribution with mean 5 and std 2.

2. Step 2: Use cdf to find probability less than 7

   Calling cdf(pd, 7) returns the cumulative probability up to 7.

3. Step 3: Check other options

   p = normcdf(7, 5, 2); uses normcdf which is valid but not using makedist. pd = makedist('Normal', 5, 2); p = pdf(pd, 7); incorrectly uses pdf instead of cdf. p = cdf('Normal', 7, 5, 2); has wrong syntax for cdf.

4. Final Answer:

   pd = makedist('Normal', 'mu', 5, 'sigma', 2); p = cdf(pd, 7); -> Option A

5. Quick Check:

   Use makedist + cdf for cumulative probability [OK]

Hint: Use cdf with makedist object for cumulative probability [OK]

Common Mistakes:

• Using pdf instead of cdf
• Wrong syntax for cdf
• Ignoring distribution parameters