MechanicsX-lecture

Courseware and learning materials

Datasets preparation

A download.sh script is provided to download the MNIST dataset into the mnist/ folder. Run it from the repository root:

source download.sh

If you encounter any issue with dataset downloading during the first run of the demos, you can manually move the files in the assets/ directory to the mnist/ folder.

Environment setup

This repository is a Julia project. The Project.toml declares all dependencies required to run the demos.

Prerequisites:

• Julia (tested with version 1.12)
• Git (to clone the repo)
• Internet access (optional, to download the MNIST dataset)

Get the sources and instantiate the environment:

# Clone and enter the repo
git clone https://github.com/CFDML/MechanicsX-lecture.git
cd MechanicsX-lecture

# Start Julia with this project and install dependencies
julia --project=.

# In the Julia REPL (Pkg mode)
# press ] to enter pkg mode, then run:
(MechanicsX-lecture) pkg> instantiate
(MechanicsX-lecture) pkg> status

Run the MNIST demos

The mnist/ folder contains two runnable examples:

• mnist/mlp.jl: a simple multilayer perceptron working on flattened 28×28 images.
• mnist/conv.jl: a small CNN (LeNet-style) operating on 28×28 images.

Both scripts will:

• cd into their own directory, so all paths are relative to mnist/.
• Download the MNIST dataset into mnist/ on first run using MLDatasets.
• Train for a small number of epochs and print periodic metrics.
• Render a sample image using ImageCore.

Simply run from the repo root:

# MLP demo
julia --project=. mnist/mlp.jl

# CNN demo
julia --project=. mnist/conv.jl

Afterwards, you are encouraged to run the scripts using Visual Studio Code with the Julia extension. You may modify hyperparameters, model architecture, and training loops to experiment with different settings.

What to expect on first run

• Terminal image not displayed: ensure your terminal supports true color; you can comment out the visualization lines without affecting training.
• Version conflicts: run pkg> instantiate again to re-sync with Project.toml.
• The first epoch may be a bit slower due to JIT compilation.

License

See LICENSE in this repository.