Video summary
Data Science Websites Most People Don’t Know Exist
Main summary
Key takeaways
Main Ideas / Lessons
- Most people rely on Google for data science learning, but several lesser-known websites can save time, provide ready-to-use datasets, and make math-heavy concepts easier to understand through interactivity or visuals.
- A recurring theme is that good resources either:
- Reduce friction in getting data (dataset repositories/newsletters),
- Reduce friction in understanding concepts (interactive learning tools, visual explanations), or
- Help practice/communicate results (public dashboards and visualizations).
Key Resources Mentioned (What They Do and Why They Matter)
Google Dataset Search
- A search engine specifically for finding datasets (separate from regular Google).
- Lets you type a topic (e.g., air quality) and retrieves datasets from government agencies, universities, and research groups.
- Includes filters such as:
- File type
- Availability
- Main benefit: avoids visiting many websites to find usable data for a project.
UCI Machine Learning Repository
- A long-standing collection of machine-learning-ready datasets (described as predating Kaggle).
- Main benefit: datasets are already available, so you don’t spend time scraping or cleaning messy data before starting.
- Example:
- Iris dataset: download it, load it into code, and begin experimenting with ML models quickly.
Data is plural (weekly newsletter + searchable dataset archive)
- A curated, searchable archive of interesting datasets.
- Examples mentioned:
- Historical shipwrecks
- Jeans pocket size (dataset category)
- Curated by Jeremy Singer-Vine, with new datasets gathered weekly.
- Main benefit: helps with the harder-than-analysis part—finding interesting data, especially for personal projects and portfolio work.
Seeing Theory
- An interactive learning tool for probability and statistics.
- Focuses on learning by interaction rather than staring at formulas.
- How it works:
- Users move sliders
- Change values
- Watch outcomes/results update
- Example concept:
- Central Limit Theorem: watching distributions change helps concepts “click” more effectively than rereading formulas.
R2D3
- Designed for people who are completely new to machine learning.
- Teaches ML through a visual story and explanations (no coding required).
- Example walkthrough:
- Predict whether a house is in San Francisco or New York
- Covers concepts such as:
- Decision trees
- Overfitting
- How a model makes decisions
- Main benefit: build intuition first before math/code.
Setosa.io
- Interactive explanations for topics commonly taught with heavy equations.
- Covers areas such as:
- Principal component analysis (PCA)
- Markov chains
- Image kernels
- Example (PCA):
- Users can move data points and observe how principal components change.
- Main benefit: understanding through interaction rather than memorizing formulas.
Distill
- Explains machine learning research using interactive visuals and diagrams.
- Value proposition: avoids struggling through dense academic papers.
- Important status note:
- Currently on an indefinite hiatus (no new articles),
- but the existing archive remains available.
- Main benefit: easier comprehension of ML research concepts without reading a long PDF.
OpenIntro
- For learning statistics from scratch using free, college-level materials.
- Includes:
- Free statistics textbooks
- Videos
- Practice problems
- Datasets
- Main benefit: build a free introductory statistics course by downloading and working through exercises with provided datasets.
Tableau Public
- A large public gallery of data visualizations.
- Users publish:
- Dashboards
- Maps
- Charts
- Interactive reports
- Main benefit: browse real-world visualization examples created by others.
Methodologies / Instructions Presented
Use Google Dataset Search to find datasets efficiently
- Search for a topic (e.g., “air quality”).
- Use filters to narrow results (e.g., by file type and availability).
- Pull datasets from many sources (government, universities, research groups) from one place.
- Use the found dataset directly for data science projects to save time.
Use interactive learning sites to make concepts “click”
- Seeing Theory
- Identify a concept you’re stuck on (example: Central Limit Theorem).
- Interact by changing values using sliders.
- Observe the distribution/results change in real time to build intuition.
- R2D3
- Learn ML via a guided visual scenario (example: SF vs NY house prediction).
- Follow how decision trees and overfitting are illustrated step-by-step.
- Avoid writing code until concepts make sense.
- Setosa.io
- For math-heavy topics (example: PCA), interact with the underlying data/components (move points).
- Observe how outputs/components change to replace memorization with intuition.
- Distill
- Use interactive diagrams/visuals to understand research ideas instead of tackling dense papers.
Build foundational learning with structured free materials
- OpenIntro
- Download free textbooks and course materials.
- Work through videos and practice problems.
- Use provided datasets to complete exercises and reinforce learning.
Learn through examples and outputs
- Tableau Public
- Browse real dashboards/reports/interactive visualizations.
- Use them as inspiration or reference for how data can be presented.
Speakers / Sources Featured
- Jeremy Singer-Vine (curator of Data is plural newsletter/dataset archive)
- Brown University students (creators of Seeing Theory)
- Beyond Google (the channel/host referenced: “Subscribe to Beyond Google…”)