Welcome to our comprehensive machine learning series 👋! The goal of this series is to give you a deeper and intuitive understanding of important concepts in machine learning. Our guides are filled with simple but insightful examples, which is a style of teaching we strongly believe in.

We aim to publish a new guide per week, and we also routinely improve our existing guides with more examples and sections.

mail

Join our newsletter for updates on new DS/ML comprehensive guides (spam-free)

memory

Machine learning models

Machine learning (ML) models are algorithms that learn from data to achieve tasks such as predictions, classification and pattern recognition. Although ML models can easily be implemented using libraries such as scikit-learn, we recommend that you take the time to understand the underlying mathematics so that you understand each model's strengths and weaknesses.

Simple Linear Regression

The "hello world of ML" that draws a line of best fit for value estimation.

Logistic Regression

A linear classifier that predicts the probability of a binary event occurring.

k-means Clustering

A clustering technique to group data points into a predefined number of clusters, k.

DBSCAN

A clustering technique that relies on density to group data points.

Hierarchical Clustering

A clustering technique to group data points into nested clusters.

Naive Bayes

A simple but powerful model based on the Bayes' theorem used often to classify text-based data.

Decision Trees

A tree-based model where the data is continuously split into a series of if-else logic.

k-Nearest neighbors

A classifier/regressor that computes predictions for a new data point based on the target values of its neighboring points.

Perceptrons

A binary linear classifier that serves as the building block of modern neural networks.

settings_suggest

Feature engineering

Data in the real world is messy and must be cleaned before model training. No matter how advanced a model is, training with low-quality data will always yield low-quality output. Feature engineering is about transforming the features to not only align their format to what the model expects, but also to improve the performance and training speed of the model.

Feature Scaling

An important preprocessing step that removes the scale of features to help build ML models.

Grid Search

A brute-force technique that finds the optimal hyper-parameters by trying out different combinations of hyper-parameters.

Random Search

A technique to determine hyper-parameter values that optimize the performance of machine learning models by random sampling.

Text and Categorical Encoding

Different approaches to encoding non-numeric data into numeric vectors that can then be fed into a ML model.

Principal Component Analysis

A dimensionality reduction technique that projects data points into lower dimensions while minimizing information loss.

Log transformation

Applies the logarithm function of some base on a set of right-skewed values so that they resemble a normal distribution.

trending_down

Optimization

Most ML models work by minimizing a particular cost function, which is simply a mathematical expression that represents how poor a model is performing. In fact, the "learning" in "machine learning" is nothing more than minimizing the cost function through some optimizer such as gradient descent.

Gradient Descent

An iterative algorithm used to determine the value of inputs that minimize a particular function (e.g. cost functions).

assignment_turned_in

Evaluating machine learning models

Once we build a ML model, we must evaluate its performance to check whether it aligns with our expectations. The evaluation metric depends on the nature of our task (e.g. classification or regression). We can also compare the performance of different ML models, and keep the best-performing model.

Confusion Matrix

A simple table used to summarize the performance of a classification algorithm.

ROC Curves

A way to visualize the performance of a binary classifier as the classification threshold changes.

Cross-validation

A technique that measures the performance of a model through resampling to reduce bias.

Mean Squared Error

A performance metric of how well our ML model fits the target numeric values.

Mean Absolute Error

A performance metric defined as the average of all the absolute differences between true and predicted values.

Root Mean Squared Error

A performance metric defined as the square root of the average squared differences between true and predicted values.

R-squared

A performance metric for a linear regression model's goodness-of-fit.

grade

PySpark

Spark is an open-source computing framework designed to handle big data. The basic idea is to split up the data into smaller partitions, and place them on multiple machines to facilitate parallel computing. PySpark is a wrapper language for Spark that allows us to interact with the framework using Python.

Getting Started with PySpark

Spark is an open-source distributed computing framework to handle big data.

Resilient Distirbuted Data

RDD is the core data structure of Spark where the data is partitioned across multiple worker nodes.

Getting Started with Databricks

Databricks offer a platform to gain some hands-on experience with PySpark for free using the community edition.

Caching RDDs and DataFrames

Caching is vital for optimization as Spark will reuse the cached RDDs and DataFrames instead of recomputing them.

User-defined functions

Using SQL to query data

Spark allows querying for data using SQL expressions instead of using built-in methods.