ML for Beginners is Microsoft's free 12-week course in classic machine learning, with 26 lessons and 52 short quizzes. It sticks to methods that run on any laptop, using Scikit-learn instead of deep learning: regression, classification, clustering, natural language processing, time series forecasting and reinforcement learning, plus two closing lessons on real-world uses and on debugging models. In the first ten minutes you can read lesson one in your browser and take its warm-up quiz, with nothing installed.
Each lesson has a written guide, a Jupyter notebook to work through, a finished solution, an assignment and a challenge, and some have short videos. The projects use real data from around the world, such as North American pumpkin prices, Asian and Indian cuisines, Nigerian music and European hotel reviews, and two of them turn a trained model into a small web app. Nine lessons also come in R, and the course is translated into more than 50 languages. Deep learning is left to Microsoft's separate AI for Beginners course.
It is written by Microsoft Cloud Advocates, including Jen Looper, and is MIT-licensed, with about 92,400 stars and 158 contributors since March 2021. There are no versioned releases; the lessons are updated in place, most recently on 6 November 2026.
- Repository: github.com/microsoft/ML-For-Beginners
- Licence: MIT (MIT License)
- Language: Jupyter Notebook. Stars: 92.4K. Forks: 22.9K. Last push: Nov 7, 2026.
- Scan: safe, Nov 21, 2026, commit 8e5c0df
Who it is for
Students, career changers and developers who want a structured first course in machine learning, and teachers who want ready-made lessons and quizzes (there is a guide for teachers). Some Python, or R, helps; the README suggests GitHub Codespaces if you would rather not install anything.
Getting started
1. Read the course on GitHub, starting with the introduction
https://github.com/microsoft/ML-For-Beginners/tree/main/1-Introduction2. Take the pre- and post-lesson quizzes online
https://ff-quizzes.netlify.app/en/ml/3. Or clone it without the 50-plus translations, to keep the download small
git clone --filter=blob:none --sparse https://github.com/microsoft/ML-For-Beginners.git4. Then, inside the folder, skip the translation folders
git sparse-checkout set --no-cone '/*' '!translations' '!translated_images'5. Install the main lesson packages and open the notebooks
pip install jupyter scikit-learn pandas numpy matplotlib seaborn && jupyter notebookA full clone with every translation is about 17,400 files and 1.2 GB; the sparse clone above checks out about 450. A few notebooks install their own extra packages, such as statsmodels, skl2onnx and gym. The quiz app can be deployed to Azure, but you do not need to: the hosted quizzes above work without it.
Safety scan
We cloned microsoft/ML-For-Beginners at commit 8e5c0df on Nov 21, 2026 and ran the checks described on the GitHub Tools page: credential patterns, decode-and-execute code, install-time scripts, committed binaries, risky CI workflows, every host the code talks to, known vulnerabilities in pinned dependencies, and project hygiene. A person read every hit. This is what we found.
- 0 secret hits across 17,444 files and about 2 million lines, most of them translated lessons and notebooks. The scanner reports no binaries; the data files worth naming are ufo-model.pkl (988 bytes) and model.onnx (2.2 MB), the finished models for the two web-app lessons, and PDFs of the course. We read the pickle's contents without loading it: it holds only a Scikit-learn LogisticRegression and NumPy arrays.
- 0 pattern hits. The lesson code has no shell calls or eval. The Flask lesson loads its model with pickle.load, the usual way Scikit-learn models are saved; pickle files can carry code, so load only the course's file or one you made yourself. A few notebooks run pip install for packages such as seaborn, statsmodels, skl2onnx and gym.
- No install hooks or install scripts. You read the lessons on GitHub or open the notebooks in Jupyter. The only npm projects are the quiz app and a tool that rebuilds the course PDF, and you need neither to learn.
- 97 advisories, 1 critical and 40 high, all in development tools, not lesson code. The critical one and most of the high ones are in the root package-lock.json, which holds docsify-to-pdf 0.0.5 and its old copies of DOMPurify, marked, extract-zip and minimatch, used only to rebuild the PDF. The rest are in quiz-app/package-lock.json, the Vue CLI build and dev-server tools; the quiz app's own Vue packages have none. The Python lessons do not pin versions, so there is nothing to look up there.
- The hosts are mostly documentation: GitHub, the tidyverse and tidymodels sites for the R lessons, Wikipedia, Scikit-learn and Microsoft Learn, whose links carry a WT.mc_id tag Microsoft uses to count course traffic. We found no analytics code. Three workflows (locking closed issues, closing stale ones, and a provenance template that runs only on releases, which the repository has none of), no pull_request_target, and both third-party actions are pinned to commits. Security policy, Dependabot for actions, contributing guide and code of conduct present; no CodeQL.
What the scanner counted
| Check | Result |
|---|---|
| Secrets | None found. |
| Suspicious code | None found. |
| Install-time code | None: nothing runs at install beyond the package manager itself. |
| Committed binaries | None. |
| CI workflows | 3 workflows. None use pull_request_target. 0 of 2 third-party actions pinned to a tag rather than a commit. |
| Network hosts | 40 distinct hosts referenced from source; most often github.com, www.tidymodels.org, en.wikipedia.org, www.tidyverse.org. No URLs to bare IP addresses. |
| Known vulnerabilities | 97 advisories across 1,044 pinned packages: 1 critical, 40 high, 45 moderate, 11 low. package-lock.json: 272 packages, 62 advisories; quiz-app/package-lock.json: 992 packages, 36 advisories. |
| Project hygiene | Has security policy, automated dependency updates, licence file, contributing guide. Missing CodeQL. |
| OpenSSF Scorecard | Not scored: the project is not in Scorecard's weekly index. |
The raw findings
Every hit the scanner wrote out, with a link to the exact line at the scanned commit. Secrets candidates are redacted.
Worst known vulnerabilities (24 of 97)
| Advisory | Severity | Package | Summary |
|---|---|---|---|
| GHSA-p3vf-v8qc-cwcr | critical | dompurify@2.2.9 | DOMPurify vulnerable to tampering by prototype polution |
| GHSA-93q8-gq69-wqmw | high | ansi-regex@4.1.0 | Inefficient Regular Expression Complexity in chalk/ansi-regex |
| GHSA-3jxr-9vmj-r5cp | high | brace-expansion@1.1.11 | brace-expansion: DoS via exponential-time expansion of consecutive non-expanding {} groups |
| GHSA-6j4f-fj2g-mc7p | high | brace-expansion@1.1.11 | brace-expansion: DoS via uncontrolled recursion in parseCommaParts causing stack exhaustion |
| GHSA-mh99-v99m-4gvg | high | brace-expansion@1.1.11 | brace-expansion: DoS via unbounded expansion length causing an out-of-memory process crash |
| GHSA-qhr7-859c-m2p7 | high | brace-expansion@1.1.11 | brace-expansion: DoS via uncontrolled recursion on nested brace groups causing stack exhaustion |
| GHSA-rgw5-rvv9-x895 | high | brace-expansion@1.1.11 | brace-expansion: DoS via unbounded intermediate arrays, bypassing the CVE-2026-14257 mitigation |
| GHSA-grv7-fg5c-xmjg | high | braces@3.0.2 | Uncontrolled resource consumption in braces |
| GHSA-gx9m-whjm-85jf | high | dompurify@2.2.9 | DOMpurify has a nesting-based mXSS |
| GHSA-mmhx-hmjr-r674 | high | dompurify@2.2.9 | DOMPurify allows tampering by prototype pollution |
| GHSA-7pqw-9j4j-h8q3 | high | extract-zip@1.7.0 | extract-zip allows arbitrary file writes through symlink archive entries |
| GHSA-jmr9-qjv8-65gv | high | extract-zip@1.7.0 | extract-zip unvalidated symlink path traversal |
| GHSA-62ch-8vmq-8xm7 | high | figlet@1.5.0 | figlet is vulnerable to denial of service via unbounded loop when whitespaceBreak is used with a small width |
| GHSA-rc47-6667-2j5j | high | http-cache-semantics@4.1.0 | http-cache-semantics vulnerable to Regular Expression Denial of Service |
| GHSA-2883-xcg3-v3hh | high | js-yaml@3.14.1 | js-yaml: maxTotalMergeKeys does not limit CPU use for empty merge sources |
| GHSA-52cp-r559-cp3m | high | js-yaml@3.14.1 | js-yaml: YAML merge-key chains can force quadratic CPU consumption |
| GHSA-5p4m-2wfm-xmqj | high | js-yaml@3.14.1 | JS-YAML: Quadratic CPU consumption in !!omap resolution (3.x and 4.x) - CVE-2026-59870 fix not backported |
| GHSA-9c47-m6qq-7p4h | high | json5@0.5.1 | Prototype Pollution in JSON5 via Parse Method |
| GHSA-r5fr-rjxr-66jc | high | lodash@4.17.21 | lodash vulnerable to Code Injection via `_.template` imports key names |
| GHSA-5v2h-r2cx-5xgj | high | marked@2.1.1 | Inefficient Regular Expression Complexity in marked |
| GHSA-rrrm-qjm4-v8hf | high | marked@2.1.1 | Inefficient Regular Expression Complexity in marked |
| GHSA-5v2h-r2cx-5xgj | high | marked@1.2.9 | Inefficient Regular Expression Complexity in marked |
| GHSA-rrrm-qjm4-v8hf | high | marked@1.2.9 | Inefficient Regular Expression Complexity in marked |
| GHSA-23c5-xmqv-rm74 | high | minimatch@3.0.4 | minimatch ReDoS: nested *() extglobs generate catastrophically backtracking regular expressions |
By the numbers
| Stars | 92.4K |
|---|---|
| Forks | 22.9K |
| Contributors | 158 |
| Commits | 2,457 |
| Open issues | 6 |
| Open pull requests | 22 |
| Releases | 0 |
| Latest release | none tagged |
| Licence | MIT |
| Main language | Jupyter Notebook |
| Project age | 5 years |
| Last push | Nov 7, 2026 |
| Tracked files | 17,444 |
| Lines of code | 2M |
| Checkout size | 1.2 GB |
Lines by language: Jupyter 1.3M, Markdown 558.3K, JSON 107.4K, HTML 3,567, Python 1,383, YAML 233.
Questions
Is ML for Beginners free?
Yes. The lessons, notebooks and quizzes are free and MIT-licensed, and the course needs no paid service. Everything runs on your own computer, or in GitHub Codespaces, which has a free monthly allowance.
Does it teach deep learning or ChatGPT-style AI?
No, on purpose. It covers classic machine learning with Scikit-learn. Microsoft's AI for Beginners course covers neural networks, and Generative AI for Beginners, also on this list, covers building apps with large language models.
Do I need to know how to code?
Some Python helps, since the lessons are Jupyter notebooks; nine lessons also come in R. You can read the lessons and take the quizzes without coding, but the projects are where most of the learning happens.
What hardware do I need?
Any ordinary laptop. The course uses classic models on small datasets, which train quickly on a normal processor, so no GPU is needed.
This post is part of GitHub Tools, where every repository is cloned and scanned before it is written up. The scan is a snapshot of one commit on one day; the repository has moved on since, so check it before you install.
