What Order Should
I Take Your Courses In?

Generative AI

• Nearly no code, just installing or using tools
• Should be comfortable installing and using Python libraries and notebooks

Generative AI

• Easy Python coding, just using an API
• Geared toward those who want to apply LLMs to their own projects
• Handles text, images, and video (sound coming later)

Computer Vision

• More technical, but you can skip that stuff if you're not ready (e.g. object detection), and just focus on the Stable Diffusion part (generating images)
• Use KerasCV, Python, Tensorflow, PyTorch, & JAX for Image Recognition, Object Detection, and Stable Diffusion
• Learn to train your own object detector
• Learn how to build your own object detection dataset

Natural Language Processing

• Also a highly technical course, but you can skip the advanced sections and focus only on the one-liners if you just want practical applications
• Massive course split into beginner, intermediate, and advanced sections
• Beginner: focus on Hugging Face plug-and-play library with pretrained models and pipelines
• Many applications including sentiment analysis, text summarization, text generation, question-answering
• Intermediate: fine-tuning Hugging Face models with applications to text classification and sequence classification
• Advanced: build transformers from scratch

Math

• Optional: take only if you need to review or you don't recognize the course content (don't just go by the title, check the course descriptions and curricula)
• Not a single course - see all courses listed below
• Calculus: differentiation and partial differentation are the most relevant
• Linear Algebra: matrices, vectors, inner products, rank, inverse, determinant, linear systems, eigenvalues
• Matrix Calculus: apply calculus to vectors and matrices (these are the "objects" we deal with in machine learning and AI)
• Probability: random variables, PMFs, PDFs, CDFs, law of large numbers, central limit theorem, generating functions

Programming

• Learn how to apply the math you learned in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Machine Learning

• Linear models for classification and regression form the basis of neural networks
• Start with linear regression, since there is a closed-form solution

Machine Learning

• Logistic regression is a model of the "neuron"
• The neuron is the building block for neural networks

Deep Learning

• "Deep learning" is really just neural networks. Learn about how neural networks really work, especially the backpropagation algorithm
• This course heavily depends on your calculus skills

Deep Learning

• Learn about modern deep learning techniques (momentum, RMSprop, Adam, batch norm, dropout, mini-batch gradient descent, how to use the GPU to accelerate computation, and modern deep learning libraries)

Deep Learning

• Apply the deep learning techniques you've learned to convolutional neural networks - neural networks specially designed for processing images.

Deep Learning

• Gain further insight with unsupervised deep learning. Very useful for understanding how the intermediate layers of a neural network give you "latent representations" of the input image / text / etc.

Deep Learning

• Apply the deep learning techniques you've learned to sequential data (time series and text)
• Simple recurrent unit, LSTM, GRU
• What happens if you try to predict stock prices? What are so many others doing wrong?

Deep Learning

• Take a short break from "in-depth" study and learn about Tensorflow 2 and PyTorch, the most popular modern deep learning libraries today. These will give you a condensed look at several application areas, including NLP, recommender systems, GANs, and reinforcement learning. These topics will also be covered more in-depth later on this path.
• Tensorflow 2 is centered around the Keras API, which is a high-level API that makes a lot of deep learning super easy

Deep Learning

• PyTorch is the deep learning library you eventually want to learn
• Used by major players in deep learning, such as OpenAI, Facebook / Meta, etc.
• Keras supports PyTorch as a backend, so you can combine what you learned in this course with what you learned previously
• There are several application areas you can branch out to, and they don't necessarily have to be done in sequence (i.e. there's no reason you need to cover computer vision before NLP, or vice versa). So the below ordering is flexible.
• The main application areas are: computer vision, NLP, reinforcement learning, and recommender systems.

Computer Vision

• Apply the previous insights to the modern technique of transfer learning. Learn to apply it to other computer vision applications, like object detection and neural style transfer.

Computer Vision

• Use KerasCV, Python, Tensorflow, PyTorch, & JAX for Image Recognition, Object Detection, and Stable Diffusion
• Learn to train your own object detector
• Learn how to build your own object detection dataset
• Unlike the earlier computer vision course, this one is entirely practical - no real math, just how to do stuff

Computer Vision

• Learn how neural networks and computer vision can be used to not just analyze images, but generate them too.
• GANs were used long before diffusion models and also work much faster.

Natural Language Processing

• How deep learning can be used with text
• Vector / bag of words representations of text and text preprocessing
• Information retrieval systems
• Probabilistic sequence models, language models
• Machine learning approaches (both supervised and unsupervised)
• Deep learning (neural network) approaches

Natural Language Processing

• If you took "Data Science: Natural Language Processing in Python", note that it has been superseded by "Machine Learning: Natural Language Processing in Python (V2)". If you are new, just take V2. Of course, if you want to take both, I am always grateful for your support.

Natural Language Processing

• In-depth into word embeddings: word2vec and GloVe
• Code them from scratch! (without autodiff libraries like Tensorflow / PyTorch, but we use those as well)
• POS tagging and NER with sequential models
• Recursive neural networks and recursive neural tensor networks (RNTNs) (i.e. tree neural networks)

Natural Language Processing

• Dives deeper into sequential models with bidirectional RNNs and seq2seq
• Introduces the all-important "attention" mechanism (precursor to transformers)
• Neural machine translation
• Memory networks

Natural Language Processing

• Massive course split into beginner, intermediate, and advanced sections
• Beginner: focus on Hugging Face plug-and-play library with pretrained models and pipelines
• Many applications including sentiment analysis, text summarization, text generation, question-answering
• Intermediate: fine-tuning Hugging Face models with applications to text classification and sequence classification
• Advanced: build transformers from scratch

Data Science

• Remember this course can be taken at any time after you know the basics! (after you learn Keras/Tensorflow 2)
• Matrix factorization will apply what you learned in linear regression
• We'll also apply Restricted Boltzmann Machines (RBMs) (unsupervised deep learning)

Reinforcement Learning

• This course doesn't contain any deep learning, but it's a prerequisite for deep reinforcement learning, which does
• Bandit algorithms
• Markov Decision Processes (MDPs), Bellman Equations
• Policy Iteration, Value Iteration
• Basic approaches to RL: Dynamic Programming (DP), Monte Carlo (MC) Methods, and Temporal Difference (TD) Learning
• Apply function approximation to RL - later of course, the "function approximator" will be a deep neural network

Reinforcement Learning

• How to make reinforcement learning "work" when you incorporate neural networks
• DQN / Deep Q-Learning
• Policy Gradient, Actor-Critic methods
• REINFORCE, A2C
• This is the updated version of this course

Reinforcement Learning

• Continue learning about deep reinforcement learning methods
• Evolutionary algorithms ("evolution strategies")
• A2C (an Actor-Critic method related to A3C)
• DDPG (Deep Deterministic Policy Gradient)
• More complex environments (Flappy Bird, MuJoCo, Atari games)

Programming

• This path doesn't focus on math, but deep learning is mathematical in nature, so while you won't learn math here, you will apply it, as it doesn't make sense to remove math entirely
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Deep Learning

• Tensorflow 2 is centered around the Keras API, which is a high-level API that makes a lot of deep learning super easy
• This will take you from beginner to advanced with applied deep learning. There is some math in the course, but this is for intuition only and can be considered optional (it's only intuitive if you actually know math and have some familiarity with machine learning basics). You'll start with linear models, then feedforward neural networks, then specialized networks like convolutional neural networks and recurrent neural networks. Through this process, you'll learn how to deal with tabular data, image data, time series data, and text data.
• After learning about the basic architectures, you'll then apply those neural networks to various areas of application, like NLP, recommender systems, GANs, computer vision with transfer learning, and deep reinforcement learning for stock trading.
• While this may seem like a large amount of topics, note that it is condensed. It is not meant to give you an in-depth view (that's what the "in-depth program" is for!), but rather, a brief look at a wide array of applications. As an example, in this course, linear regression is half a section, while in the "in-depth program" it comprises an entire course.
• If at any point, you find yourself wondering, "but how does this actually work?", then you are a good candidate for the in-depth series.

Deep Learning

• PyTorch is the deep learning library you eventually want to learn
• Used by major players in deep learning, such as OpenAI, Facebook / Meta, etc.
• Keras supports PyTorch as a backend, so you can combine what you learned in this course with what you learned previously
• There are several application areas you can branch out to, and they don't necessarily have to be done in sequence (i.e. there's no reason you need to cover computer vision before NLP, or vice versa). So the below ordering is flexible.
• The main application areas are: computer vision, NLP, reinforcement learning, and recommender systems.

Computer Vision

• Apply the previous insights to the modern technique of transfer learning. Learn to apply it to other computer vision applications, like object detection and neural style transfer.

Computer Vision

• Use KerasCV, Python, Tensorflow, PyTorch, & JAX for Image Recognition, Object Detection, and Stable Diffusion
• Learn to train your own object detector
• Learn how to build your own object detection dataset
• Unlike the earlier computer vision course, this one is entirely practical - no real math, just how to do stuff

Natural Language Processing

• How deep learning can be used with text
• Vector / bag of words representations of text and text preprocessing
• Information retrieval systems
• Probabilistic sequence models, language models
• Machine learning approaches (both supervised and unsupervised)
• Deep learning (neural network) approaches

Natural Language Processing

• Dives deeper into sequential models with bidirectional RNNs and seq2seq
• Introduces the all-important "attention" mechanism (precursor to transformers)
• Neural machine translation
• Memory networks

Natural Language Processing

• Massive course split into beginner, intermediate, and advanced sections
• Beginner: focus on Hugging Face plug-and-play library with pretrained models and pipelines
• Many applications including sentiment analysis, text summarization, text generation, question-answering
• Intermediate: fine-tuning Hugging Face models with applications to text classification and sequence classification
• Advanced: build transformers from scratch

Programming

• Learn the basics of numerical and scientific computing in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Machine Learning

• The next step is to learn about simple linear models for supervised learning (both classification and regression). "Simple" in this case does not imply "less useful". In fact, linear models will probably be the most used, due to their simplicity and importantly, their interpretability. This will be extremely valuable when pitching your technology to stakeholders.

Machine Learning

• Again, very practical and interpretable. Example use-cases: predict whether or not the user will purchase an item, predict the amount of sales given adspend, detect fraudulent behavior on your website.

Machine Learning

• Before moving on to neural networks, get to know other, more interesting supervised machine learning models.
• K-Nearest Neighbor, Decision Trees, Perceptron, Naive Bayes
• How to deploy your model to a server / API web service!

Machine Learning

• Naive Bayes, covered in much more detail than the Supervised Machine Learning course.
• I am working on replacing that course.

Machine Learning

• Learn how to combine decision trees (and other ML models) to create even more powerful models called ensembles
• Covers an important ML/statistical learning concept: the Bias-Variance tradeoff
• 2 essential approaches: Random Forest and AdaBoost

Machine Learning

• Let's now explore the world of unsupervised learning (for data without labels)
• Can be taken prior to or in parallel with above supervised ML courses
• Business people love clustering
• K-means clustering, hierarchical clustering, and Gaussian mixture models
• Expectation-maximization algorithm (applied again later with HMMs)

Statistics

• One of the most practical techniques for business: A/B Testing. Is advertisement A better than advertisement B? Is sales copy A better than sales copy B? Answer these questions quantitatively. Additionally, the more advanced techniques you will learn will allow you to build adaptive systems that "learn" on-the-fly.
• Extremely useful for online business, online advertising, and online marketing.

Data Science

• Will apply the Bayesian bandit method again, but this time in the context of sorting feeds (e.g. news feeds, Instagram feeds, ...)
• Another extremely practical set of algorithms used by all the biggest tech companies, such as Google, Facebook, Netflix, Amazon, etc. - recommender systems. Whenever you display a list of products to the user, you'd like to do so intelligently. Show users the products they are most likely to purchase or rate highly.

Math

• Optional: take only if you need to review or you don't recognize the course content (don't just go by the title, check the course descriptions and curricula)
• Not a single course - see all courses listed below
• Calculus: differentiation and partial differentation are the most relevant
• Linear Algebra: matrices, vectors, inner products, rank, inverse, determinant, linear systems, eigenvalues
• Matrix Calculus: apply calculus to vectors and matrices (these are the "objects" we deal with in machine learning and AI)
• Probability: random variables, PMFs, PDFs, CDFs, law of large numbers, central limit theorem, generating functions
• Note: you need to be strong in probability to understand reinforcement learning

Programming

• Learn how to apply the math you learned in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Machine Learning

• Linear models for classification and regression form the basis of neural networks
• Start with linear regression, since there is a closed-form solution

Machine Learning

• Logistic regression is a model of the "neuron"
• The neuron is the building block for neural networks

Deep Learning

• "Deep learning" is really just neural networks. Learn about how neural networks really work, especially the backpropagation algorithm
• This course heavily depends on your calculus skills

Deep Learning

• Learn about modern deep learning techniques (momentum, RMSprop, Adam, batch norm, dropout, mini-batch gradient descent, how to use the GPU to accelerate computation, and modern deep learning libraries)

Reinforcement Learning

• Take a break from deep learning, learn the basics of RL
• Bandit algorithms
• Markov Decision Processes (MDPs), Bellman Equations
• Policy Iteration, Value Iteration
• Basic approaches to RL: Dynamic Programming (DP), Monte Carlo (MC) Methods, and Temporal Difference (TD) Learning
• Apply function approximation to RL - later of course, the "function approximator" will be a deep neural network

Deep Learning

• Take a short break from "in-depth" study and learn about Tensorflow 2 and PyTorch, the most popular modern deep learning libraries today. These will give you a condensed look at several application areas, including NLP, recommender systems, GANs, and reinforcement learning. These topics will also be covered more in-depth later on this path.
• Tensorflow 2 is centered around the Keras API, which is a high-level API that makes a lot of deep learning super easy

Deep Learning

• PyTorch is the deep learning library you eventually want to learn
• Used by major players in deep learning, such as OpenAI, Facebook / Meta, etc.
• Keras supports PyTorch as a backend, so you can combine what you learned in this course with what you learned previously
• There are several application areas you can branch out to, and they don't necessarily have to be done in sequence (i.e. there's no reason you need to cover computer vision before NLP, or vice versa). So the below ordering is flexible.
• The main application areas are: computer vision, NLP, reinforcement learning, and recommender systems.

Reinforcement Learning

• It's now time to apply deep learning to reinforcement learning. Here the 2 subjects converge.
• How to make reinforcement learning "work" when you incorporate neural networks
• DQN / Deep Q-Learning
• Policy Gradient, Actor-Critic methods
• REINFORCE, A2C
• This is the updated version of this course

Reinforcement Learning

• Continue learning about deep reinforcement learning methods
• Evolutionary algorithms ("evolution strategies")
• A2C (an Actor-Critic method related to A3C)
• DDPG (Deep Deterministic Policy Gradient)
• More complex environments (Flappy Bird, MuJoCo, Atari games)

Math

• Optional: take only if you need to review or you don't recognize the course content (don't just go by the title, check the course descriptions and curricula)
• Not a single course - see all courses listed below
• Calculus: differentiation and partial differentation are the most relevant
• Linear Algebra: matrices, vectors, inner products, rank, inverse, determinant, linear systems, eigenvalues
• Matrix Calculus: apply calculus to vectors and matrices (these are the "objects" we deal with in machine learning and AI)
• Probability: random variables, PMFs, PDFs, CDFs, law of large numbers, central limit theorem, generating functions

Programming

• Learn how to apply the math you learned in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Machine Learning

• Linear models for classification and regression form the basis of neural networks
• Start with linear regression, since there is a closed-form solution

Machine Learning

• Logistic regression is a model of the "neuron"
• The neuron is the building block for neural networks

Deep Learning

• "Deep learning" is really just neural networks. Learn about how neural networks really work, especially the backpropagation algorithm
• This course heavily depends on your calculus skills

Deep Learning

• Learn about modern deep learning techniques (momentum, RMSprop, Adam, batch norm, dropout, mini-batch gradient descent, how to use the GPU to accelerate computation, and modern deep learning libraries)

Deep Learning

• Apply the deep learning techniques you've learned to convolutional neural networks - neural networks specially designed for processing images.

Deep Learning

• Gain further insight with unsupervised deep learning. Very useful for understanding how the intermediate layers of a neural network give you "latent representations" of the input image / text / etc.

Computer Vision

• Apply the previous insights to the modern technique of transfer learning. Learn to apply it to other computer vision applications, like object detection and neural style transfer.

Computer Vision

• Learn how neural networks and computer vision can be used to not just analyze images, but generate them too.
• GANs were used long before diffusion models and also work much faster.

Computer Vision

• Use KerasCV, Python, Tensorflow, PyTorch, & JAX for Image Recognition, Object Detection, and Stable Diffusion
• Learn to train your own object detector
• Learn how to build your own object detection dataset
• Unlike the earlier computer vision course, this one is entirely practical - no real math, just how to do stuff

Programming

• TLearn the basics of numerical and scientific computing in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Natural Language Processing

• How deep learning can be used with text
• Vector / bag of words representations of text and text preprocessing
• Information retrieval systems
• Probabilistic sequence models, language models
• Machine learning approaches (both supervised and unsupervised)
• Deep learning (neural network) approaches

Natural Language Processing

• If you took "Data Science: Natural Language Processing in Python", note that it has been superseded by "Machine Learning: Natural Language Processing in Python (V2)". If you are new, just take V2. Of course, if you want to take both, I am always grateful for your support.
• If you already took this course in the past, move on to Tensorflow 2 instead of NLP (V2)

Deep Learning

• Take a short break from "in-depth" study and learn about Tensorflow 2 and PyTorch, the most popular modern deep learning libraries today. These will give you a condensed look at several application areas, including NLP, recommender systems, GANs, and reinforcement learning. These topics will also be covered more in-depth later on this path.
• Tensorflow 2 is centered around the Keras API, which is a high-level API that makes a lot of deep learning super easy

Deep Learning

• PyTorch is the deep learning library you eventually want to learn
• Used by major players in deep learning, such as OpenAI, Facebook / Meta, etc.
• Keras supports PyTorch as a backend, so you can combine what you learned in this course with what you learned previously
• There are several application areas you can branch out to, and they don't necessarily have to be done in sequence (i.e. there's no reason you need to cover computer vision before NLP, or vice versa). So the below ordering is flexible.
• The main application areas are: computer vision, NLP, reinforcement learning, and recommender systems.

Machine Learning

• Linear models for classification and regression form the basis of neural networks
• Start with linear regression, since there is a closed-form solution

Machine Learning

• Logistic regression is a model of the "neuron"
• The neuron is the building block for neural networks

Deep Learning

• "Deep learning" is really just neural networks. Learn about how neural networks really work, especially the backpropagation algorithm
• This course heavily depends on your calculus skills

Deep Learning

• Learn about modern deep learning techniques (momentum, RMSprop, Adam, batch norm, dropout, mini-batch gradient descent, how to use the GPU to accelerate computation, and modern deep learning libraries)

Deep Learning

• Apply the deep learning techniques you've learned to convolutional neural networks - neural networks specially designed for processing images.

Deep Learning

• Gain further insight with unsupervised deep learning. Very useful for understanding how the intermediate layers of a neural network give you "latent representations" of the input image / text / etc.

Deep Learning

• Apply the deep learning techniques you've learned to sequential data (time series and text)
• Simple recurrent unit, LSTM, GRU
• What happens if you try to predict stock prices? What are so many others doing wrong?

Natural Language Processing

• In-depth into word embeddings: word2vec and GloVe
• Code them from scratch! (without autodiff libraries like Tensorflow / PyTorch, but we use those as well)
• POS tagging and NER with sequential models
• Recursive neural networks and recursive neural tensor networks (RNTNs) (i.e. tree neural networks)

Natural Language Processing

• Dives deeper into sequential models with bidirectional RNNs and seq2seq
• Introduces the all-important "attention" mechanism (precursor to transformers)
• Neural machine translation
• Memory networks

Natural Language Processing

• Massive course split into beginner, intermediate, and advanced sections
• Beginner: focus on Hugging Face plug-and-play library with pretrained models and pipelines
• Many applications including sentiment analysis, text summarization, text generation, question-answering
• Intermediate: fine-tuning Hugging Face models with applications to text classification and sequence classification
• Advanced: build transformers from scratch

Generative AI

Generative AI: OpenAI API, ChatGPT, and GPT-4 in Python

• As you may already know, transformers are what drive the latest mainstream AIs, like ChatGPT, GPT-4, Mistral, Google's Gemini, Meta's Llama, etc. At this level, the main architecture hasn't really changed (they're still transformers), but the way we work with them is different. You won't be training your own ChatGPT. But you will be using it! The following courses are designed to teach you how to use technologies like ChatGPT effectively, focusing mainly on their APIs (rather than more general topics like prompt engineering, though we do cover that as well). The goal is to stay within the world of programming and building tools / applications, while making use of the latest state-of-the-art LLMs. Although I've placed these at the end of the ordering, this doesn't imply that you have to take these courses last. In fact, all that is required is a bit of programming knowledge. However, to fully appreciate what these LLMs are doing, you'll want to have a good handle on the aforementioned NLP topics.

Programming

• Learn the basics of numerical and scientific computing in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Statistics

• This will introduce you to Bayesian inference in an applied setting. For instance, how to estimate the click-through rate of an advertisement or the conversion rate of a sales page. The methods you will learn in this course are very practical and can be used for just about any online business. You'll learn the difference between the classical (frequentist) and Bayesian approaches. You'll learn about hypothesis testing (i.e. "Which article performs better, A or B?"). But most importantly, this course will teach you how to use Bayes' rule. The key point you will learn is that instead of making point estimates, your estimates will be expressed as probability distributions.
• The above course doesn't cover what people "traditionally" think of as machine learning (e.g. regression, classification, clustering, etc.). Before you move on to Bayesian Linear Regression, Bayesian Classification, Bayesian Clustering, etc... it is suggested that you learn about their "non-Bayesian" variants first. The "Bayesian" component adds quite a bit more difficulty in terms of math.
• As such, the following machine learning courses are strongly recommended, but not necessarily required (e.g. Bayesian linear regression doesn't strictly depend on clustering, but it does depend on linear regression).

Machine Learning

• The next step is to learn about simple linear models for supervised learning (both classification and regression). "Simple" in this case does not imply "less useful". In fact, linear models will probably be the most used, due to their simplicity and importantly, their interpretability. This will be extremely valuable when pitching your technology to stakeholders.

Machine Learning

• Again, very practical and interpretable. Example use-cases: predict whether or not the user will purchase an item, predict the amount of sales given adspend, detect fraudulent behavior on your website.

Machine Learning

• Let's now explore the world of unsupervised learning (for data without labels)
• K-means clustering, hierarchical clustering, and Gaussian mixture models
• Expectation-maximization algorithm (applied again later with HMMs)

Bayesian Machine Learning

• One area of machine learning we've only briefly touched on at this point is Bayesian Machine Learning (through the A/B Testing course). But the above course doesn't cover what people "traditionally" think of as machine learning (e.g. regression, classification, clustering, etc.). In order to apply the Bayesian approach to these models, we will look at them in roughly the same order as above (starting with linear regression, then linear classification, then unsupervised learning, etc.)

Machine Learning

• Naive Bayes, covered in much more detail than the Supervised Machine Learning course.
• I am working on replacing that course.

Bayesian Machine Learning

• Implement a "Bayesian" Bayes Classifier (different from the Bayes classifiers discussed earlier)

Programming

• Learn the basics of numerical and scientific computing in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Statistics

• Statistics will be your primary tool in both finance and time series analysis. Most techniques rely on statistics. Start with this primer.

Time Series & Finance

• Then, jump right into finance. This course is huge.
• Note that while the above course does give you a brief overview of several advanced machine learning techniques, like reinforcement learning, HMMs, GMMs, and PCA, you might want to learn about them more in depth. The below courses can help with that.

Machine Learning

• Let's now explore the world of unsupervised learning (for data without labels)
• K-means clustering, hierarchical clustering, and Gaussian mixture models
• Expectation-maximization algorithm (applied again later with HMMs)

Machine Learning

• Ok, back to unsupervised learning. Classic machine learning methods also look at sequence models. Among the most popular is the Hidden Markov Model (HMM).
• HMMs have been successfully used for speech recognition, biological sequence (e.g. DNA and RNA) analysis, and stock trading.
• Fun fact: HMMs were used early on at Renaissance Technologies (a.k.a. RenTech - then called Monemetrics), which is one of the most popular and successful quantitative hedge funds. Not only did they use HMMs, they hired Leonard Baum, one of the inventors of the famous "Baum-Welch Algorithm" used for training HMMs.
• The HMM course draws on knowledge from the Cluster Analysis course, specifically the EM (expectation-maximization) algorithm and Gaussian Mixture Models (GMMs).

Reinforcement Learning

• Learn how RL works in-depth
• Bandit algorithms
• Markov Decision Processes (MDPs), Bellman Equations
• Policy Iteration, Value Iteration
• Basic approaches to RL: Dynamic Programming (DP), Monte Carlo (MC) Methods, and Temporal Difference (TD) Learning
• Apply function approximation to RL - later of course, the "function approximator" will be a deep neural network

Time Series & Finance

• Next, enhance your knowledge of time series with Time Series Analysis. While the financial engineering course introduces you to fundamental tools like Holt-Winters and ARIMA, this course goes much more in-depth. You'll learn about more advanced models such as GARCH.

Time Series & Finance

• As an alternative to the full Time Series course, if you for some reason only want to learn about Facebook Prophet (one of the VIP sections of the Time Series course), I have an individual course on that.

Time Series & Finance

• There are more trading algorithms out there! One popular algorithm, called Pairs Trading, is explored with the help of ChatGPT.

Programming

• Learn the basics of numerical and scientific computing in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course

Machine Learning

• The next step is to learn about simple linear models for supervised learning (both classification and regression). "Simple" in this case does not imply "less useful". In fact, linear models will probably be the most used, due to their simplicity and importantly, their interpretability. This will be extremely valuable when pitching your technology to stakeholders.

Machine Learning

• Again, very practical and interpretable. Example use-cases: predict whether or not the user will purchase an item, predict the amount of sales given adspend, detect fraudulent behavior on your website.

Machine Learning

• Before moving on to neural networks, get to know other, more interesting supervised machine learning models.
• K-Nearest Neighbor, Decision Trees, Perceptron, Naive Bayes
• How to deploy your model to a server / API web service!

Machine Learning

• Naive Bayes, covered in much more detail than the Supervised Machine Learning course.
• I am working on replacing that course.

Machine Learning

• Learn how to combine decision trees (and other ML models) to create even more powerful models called ensembles
• Covers an important ML/statistical learning concept: the Bias-Variance tradeoff
• 2 essential approaches: Random Forest and AdaBoost

Machine Learning

• Let's now explore the world of unsupervised learning (for data without labels)
• Can be taken prior to or in parallel with above supervised ML courses
• Business people love clustering
• K-means clustering, hierarchical clustering, and Gaussian mixture models
• Expectation-maximization algorithm (applied again later with HMMs)

Machine Learning

• Another very powerful ML model
• Start with Linear SVM and find the optimal separating line/hyperplane (why is it more optimal than logistic regression and perceptron?)
• Extend SVM to the nonlinear case with the "kernel trick"
• Also learn how to apply Kernel Methods to other ML models, both supervised and unsupervised
• Extend SVM to multiclass case and regression case
• Another look at quadratic programming (QP) also covered in finance course

Deep Learning

• At this point, it's time to turn our attention back to neural networks and deep learning.
• "Deep learning" is really just neural networks. Learn about how neural networks really work, especially the backpropagation algorithm
• This course heavily depends on your calculus skills

Deep Learning

• Back to unsupervised learning. Aside from clustering, another common unsupervised task is dimensionality reduction. Basically, converting high dimensional vectors into lower dimensional vectors. This can be useful, for instance, in visualizing high-dimensional data. I recommend using this course to learn about PCA (principal components analysis) and autoencoders. Consider RBMs to be optional. Later, I may make a pure PCA course to replace this one.

Machine Learning

• Classic machine learning methods also look at sequence models. Among the most popular is the Hidden Markov Model (HMM). HMMs have been successfully used for speech recognition, biological sequence (e.g. DNA and RNA) analysis, and stock trading.
• The HMM course draws on knowledge from the Cluster Analysis course, specifically the EM (expectation-maximization) algorithm and Gaussian Mixture Models (GMMs).

Data Science

• Very fun and immensely practical: recommender systems. Used by all the biggest tech companies, such as Google, Facebook, Netflix, Amazon, etc. Whenever you display a list of products to the user, you'd like to do so intelligently. Use these techniques to show users the products they are most likely to purchase or rate highly.

Reinforcement Learning

• I place reinforcement learning last because it's the most challenging to both understand and implement correctly
• Bandit algorithms
• Markov Decision Processes (MDPs), Bellman Equations
• Policy Iteration, Value Iteration
• Basic approaches to RL: Dynamic Programming (DP), Monte Carlo (MC) Methods, and Temporal Difference (TD) Learning

Math

• Alright, so if you actually want to learn machine learning, you're going to have to get serious about math.
• It all starts with calculus (assuming you have finished high school math).
• This course covers what you'd learn in calculus 1 (differentiation), 2 (integration), and 3 (vector calculus).

Math

• All about vectors, matrices, and how to manipulate them
• Why? Because that's how we represent "data" in machine learning
• A table of data (rows and columns) is just a matrix
• An image is also a matrix
• Audio and time series can be considered 1-D vectors
• Covers matrix arithmetic, determinant, inverse, and other operations, rank, definiteness, eigenvalues

Math

• Learn how to differentiate functions with respect to vectors and matrices
• Also covers optimization techniques (gradient descent, Newton's method)

Math

• Probability is a mathematical tool for quantifying uncertainty
• You'll apply both calculus and linear algebra
• Used in nearly every machine learning / AI technique you'll learn about, from linear regression to neural networks to reinforcement learning
• Covers random variables, PMFs, PDFs, CDFs, joint, conditional, and marginal distributions, Bayes' rule, random vectors, generating functions, law of large numbers, central limit theorem
• Note: I separate probability from statistics, as most proper university-level courses do. Courses that don't tend to leave out too many important details. There's a statistics course later in this series.

Programming

• More on the practical side of things. How do we actually store data in a web application? No, we don't store everything in Pandas DataFrames, you silly data scientist! We use databases. The most common type of database is the relational database, some examples of which are MySQL, PostgreSQL, and SQLite.

Programming

• Learn how to apply the math you learned in Python
• It's not "just coding", it's "coding mathy things"
• Vector and matrix operations with Numpy
• Plotting and data visualization with Matplotlib
• Loading and manipulating datasets with Pandas
• More scientific computing (statistics, signal processing) with Scipy
• This prepares you for machine learning / data science / AI courses because they will all apply one or more of the above (loading in data, manipulating the data as vectors / matrices / tensors, plotting the results, ...)
• Free (but shorter) version of the course