AI for Beginners Environment.yml Dependencies: Complete Guide to the Microsoft Curriculum Stack

The AI for Beginners environment.yml defines a hybrid Conda-and-pip environment featuring PyTorch, TensorFlow, scikit-learn, Jupyter, and 15+ core packages required to run the Microsoft's open-source AI curriculum.

The environment.yml file in the Microsoft/AI-For-Beginners repository orchestrates a reproducible machine learning environment that powers every notebook in the 24-lesson curriculum. Located at the repository root, this Conda configuration combines binary-optimized packages with a pip-managed requirements file to support both deep learning frameworks (PyTorch and TensorFlow) and interactive development tools.

Core Dependencies in the AI for Beginners Environment.yml

The primary environment.yml ([environment.yml](https://github.com/microsoft/AI-For-Beginners/blob/main/environment.yml)) at lines 6-24 specifies the following foundational packages:

Interactive Development Stack

• ipykernel — Enables Jupyter kernels to execute Python code inside notebook cells
• ipython — Provides the enhanced interactive Python shell underlying notebook sessions
• ipywidgets — Supplies interactive UI elements (sliders, buttons, dropdowns) for demonstrations
• jupyter — The notebook server that launches the browser-based interactive interface

These four packages (lines 6-9) form the complete Jupyter ecosystem used throughout the curriculum's hands-on exercises.

Scientific Computing Foundation

• numpy = 1.26 (line 11) — Fundamental array-processing library; the numerical backbone of virtually every ML operation
• scipy = 1.13 (line 15) — Scientific computing tools for linear algebra, optimization, and signal processing
• matplotlib = 3.9 (line 10) — Core plotting library for visualizing datasets, training curves, and model outputs
• scikit-learn (line 14) — Classical machine learning algorithms, preprocessing utilities, and evaluation metrics

The pinned versions of NumPy, SciPy, and Matplotlib ensure compatibility across different operating systems and prevent breaking changes from upstream updates.

Deep Learning Framework

• pytorch (line 19, pytorch channel) — Core deep learning tensor library and autograd engine
• torchvision (line 21) — Computer vision datasets, image transforms, and pretrained ResNet/VGG models
• torchtext (line 20) — Text processing utilities, tokenizers, and NLP datasets
• torchdata (line 22) — Composable data-loading pipelines that integrate with torchtext and torchvision

These PyTorch ecosystem packages are sourced from the official pytorch Conda channel rather than the default channel, ensuring optimized CUDA builds for GPU acceleration where available.

Computer Vision and Data Access

• opencv (line 17, conda-forge) — OpenCV library for image loading, resizing, color space conversion, and video processing
• requests = 2.32 (line 13) — HTTP client library used by notebooks to download datasets and interact with web APIs

OpenCV from conda-forge provides precompiled binaries that avoid common pip installation issues on Windows and macOS.

Build and Package Management

• pip (line 16) — Included as a Conda package to enable the pip: block
• setuptools (line 18) — Build utilities required for compiling certain Python packages from source
• python (line 12) — Specifies Python interpreter availability (version unconstrained to use latest compatible)

Hybrid Environment Architecture: Conda + Pip

The AI for Beginners environment.yml implements a two-tier dependency strategy (lines 16-24):

pip:
  - -r requirements.txt

This pip: block delegates additional package installation to the separate [requirements.txt](https://github.com/microsoft/AI-For-Beginners/blob/main/requirements.txt) file. According to the Microsoft/AI-For-Beginners source code, this hybrid approach achieves:

• Binary reliability — Conda handles complex native dependencies (PyTorch's CUDA kernels, OpenCV's image codecs)
• Package freshness — Pip installs rapidly-evolving ML libraries (TensorFlow 2.x, Keras 3) without waiting for Conda builds
• Cross-platform consistency — Synchronized lockfiles ensure identical environments on Windows, macOS, and Linux

Creating and Activating the Environment

Execute these commands to replicate the exact environment defined in environment.yml:

# Create the environment from the curriculum definition

conda env create -f environment.yml

# Activate the environment (name defaults to project directory or explicit name)

conda activate ai4beg

# Verify key frameworks import correctly

python -c "import torch, cv2, sklearn, numpy; print(f'PyTorch {torch.__version__}, NumPy {numpy.__version__}')"

# Launch the interactive notebook server

jupyter notebook

The environment name ai4beg is defined implicitly; check environment.yml for an explicit name: field if the activation fails.

Binder-Specific Environment Variant

The repository maintains a second environment definition at [binder/environment.yml](https://github.com/microsoft/AI-For-Beginners/blob/main/binder/environment.yml) for cloud deployment. This variant:

• Pins exact versions of all packages for reproducible Binder builds
• Removes platform-specific build constraints incompatible with the Binder container runtime
• References the same requirements.txt pip dependencies

Use this file instead when launching via mybinder.org:

conda env create -f binder/environment.yml

Complete Dependency File Reference

File	Purpose	Location
environment.yml	Primary Conda environment for local development	Repository root
binder/environment.yml	Cloud-optimized variant with explicit version pins	binder/ subdirectory
requirements.txt	Pip-only packages (TensorFlow, Keras, etc.)	Repository root
README.md	Setup instructions referencing these files	Repository root

Summary

• The AI for Beginners environment.yml specifies 14 direct Conda dependencies plus a pip requirements file
• PyTorch and its ecosystem (torchvision, torchtext, torchdata) provide the primary deep learning framework
• NumPy 1.26, SciPy 1.13, and Matplotlib 3.9 are pinned for stability across the 24-lesson curriculum
• The hybrid Conda-pip architecture balances binary reliability with access to cutting-edge ML packages
• A separate Binder environment file ensures reproducible cloud deployments with exact version constraints

Frequently Asked Questions

What deep learning frameworks are included in the AI for Beginners environment?

The environment.yml installs PyTorch as its primary deep learning framework, including the full ecosystem (torchvision, torchtext, torchdata). TensorFlow and Keras are installed via the pip requirements file referenced in the pip: -r requirements.txt block at line 23-24. This dual-framework approach supports lessons using either PyTorch or TensorFlow code examples.

Why does the environment use both Conda and pip?

The Microsoft/AI-For-Beginners source code uses Conda for packages requiring complex binary compilation (PyTorch with CUDA, OpenCV image codecs) while delegating to pip for pure-Python packages that update frequently (TensorFlow 2.x, Keras 3). This hybrid method avoids common installation failures while maintaining access to the latest ML library versions.

How do I troubleshoot import errors after creating the environment?

First verify activation with conda activate ai4beg (or your specified environment name). Then check individual imports: python -c "import torch, tensorflow as tf, cv2, sklearn". If TensorFlow fails, the pip requirements.txt installation may have been skipped—run pip install -r requirements.txt manually. Platform-specific OpenCV issues usually resolve by reinstalling from conda-forge: conda install -c conda-forge opencv.

What Python version does the AI for Beginners environment use?

The environment.yml (line 12) lists python without an explicit version constraint, which allows Conda to select the latest compatible Python version for the dependency set. The pinned NumPy 1.26 and SciPy 1.13 versions currently constrain this to Python 3.9-3.11. For exact Python version requirements, check the binder/environment.yml variant which specifies explicit interpreter versions.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}