ccexam 0.0.6

Digit classifiers (MNIST / USPS / SVHN) with a unified training, evaluation and inference CLI.

ACME Digit Classification

ACME Digit Classification is a machine learning framework for handwritten digit recognition developed as part of the FYS-8805 Collaborative Coding Exam at UiT.

The repository provides a unified interface for training, evaluating, and deploying digit classification models for three customer datasets:

• Customer A: MNIST
• Customer B: SVHN
• Customer C: USPS

---

Installation

Install the latest release from PyPI:

pip install ccexam

To upgrade an existing installation to the latest release:

pip install --upgrade ccexam

Or install the latest development version directly from GitHub:

pip install git+https://github.com/FYS-8805-Collaborative-Coding/Collaborative-Coding-Exam.git

For development

Clone the repository and create the conda environment from environment.yml:

git clone https://github.com/FYS-8805-Collaborative-Coding/Collaborative-Coding-Exam.git
cd Collaborative-Coding-Exam
conda env create -f environment.yml
conda activate collaborative-coding-exam

---

Quick start: run inference

After installing the package (pip install ccexam), run inference from the command line with the ccexam-infer command:

# Classify a single image
ccexam-infer --model svhn --input datasets/inference/svhn_digit_5.png

# Classify every image in a directory
ccexam-infer --model svhn --input datasets/inference

# Try it instantly on a bundled sample (no file needed) — see "Quick test" below
ccexam-infer --model svhn --input samples:svhn_digit_5.png

# Force CPU (e.g. on a laptop with no GPU)
ccexam-infer --model svhn --input mydigit.png --device cpu

Available models: mnist, usps, svhn. The trained model weights are bundled with the package and loaded automatically — no extra setup needed.

Supported inputs

--input accepts a single file or a directory, and is flexible about formats:

• Any image, detected by content — not by extension. A real image is accepted even if it has the "wrong" extension or no extension at all (e.g. --input mydigit). A non-image is rejected with a error Invalid input: … message and a non-zero exit code.
• ASCII-art digits in .txt files. A digit drawn with characters (foreground vs. spaces) is auto-detected and converted to an image before inference — no flag needed:

  ccexam-infer --model mnist --input datasets/inference/ascii_digit_5.txt
  

• Directories skip any unreadable files (with a logged note) instead of aborting the whole run.

Quick test with bundled samples

To try the package without supplying your own files, use the samples: scheme. It resolves to an example image shipped inside the package, so it works straight after pip install:

ccexam-infer --model svhn  --input samples:svhn_digit_5.png
ccexam-infer --model mnist --input samples:mnist_test_0_label_7.png
ccexam-infer --model mnist --input samples:ascii_digit_5.txt     # bundled ASCII sample

Saving predictions to a file

By default, predictions are only printed to the terminal. Pass --output (-o) to also write them to a file:

# Write predictions to results/predictions.csv (the results/ folder is created automatically)
ccexam-infer --model svhn --input datasets/inference --output results/predictions.csv

Notes:

• The path you give is used exactly as written — any folders in it (e.g. results/) are created for you.
• Existing files are never overwritten. If results/predictions.csv already exists, the next run writes results/predictions_1.csv, then _2, and so on, so previous results are preserved.
• The format follows the file extension: .csv writes an image,prediction table with a header row; .txt writes one <image>\t<prediction> line per image.

Python API

from ccexam import run_inference

# A single image returns one label
label = run_inference(model="svhn", input_path="digit.png")
print(label)     # 5

# A folder of images returns a list of labels (sorted by filename)
labels = run_inference(model="svhn", input_path="folder_of_digits/")
print(labels)    # [7, 2, 1, 0]

When working from a repository checkout without installing, you can run the same thing as a module: python -m src.inference --model svhn --input <path> --output results/predictions.csv.

---

Model Cards

Model A — MNIST

Model A is an MNIST digit classifier for recognizing handwritten digits from 28x28 grayscale images.

Architecture	MNISTNet: small CNN with two convolution/ReLU/max-pool blocks and a fully connected classifier with 128 unit hidden dimension.
Training data	MNIST handwritten digit training set.
Intended use	Classify MNIST-like digit images into classes 0-9.
Limitations	Intended for clean MNIST style grayscale digits; performance may drop on other image styles, noise, or non digit inputs.

Performance:

Metric	Value
Precision	0.9905
Recall	0.9905
Speed (inference)	0.060 ms / sample

---

Model B — SVHN

CNN classifier for Street View House Numbers. It predicts a single cropped house-number digit (0–9) from a 32×32 RGB image.

Architecture	3-block CNN (per block: Conv-BN-ReLU ×2 + MaxPool, channels 3→32→64→128) followed by a 2-layer fully-connected head with dropout 0.3
Training data	SVHN train_32x32.mat (~73k 32×32 RGB cropped-digit images), trained 5 epochs, Adam, lr 1e-3, batch size 64
Intended use	Classifying house-number digits
Limitations	Single digits only (not multi-digit house numbers). Fixed 32×32 RGB input. Trained only 5 epochs with no data augmentation

Performance: (measured on the SVHN test set, weights/svhn.pth, LUMI-G / MI250x)

Metric	Value
Precision	0.9465
Recall	0.9465
Speed (inference)	0.470 ms / sample

---

Model C — USPS

CNN classifier trained on the USPS (United States Postal Service) dataset consisting of ~10,000 images of handwritten digits.

Architecture	USPSNet: small CNN with two convolution/ReLU/max-pool blocks and a fully connected classifier with 128 unit hidden dimension.
Training data	USPS (United States Postal Service) dataset consisting of ~10,000 16x16 grayscale images
Intended use	Classifying grayscale handwritten digits
Limitations	Limited amount of data, with fixed 16x16 grid. Only grayscale.

Performance:

Metric	Value
Precision	0.9631
Recall	0.9631
Speed (inference)	0.033 ms / sample

Documentation

More detailed documentation is available at https://fys-8805-collaborative-coding.github.io/Collaborative-Coding-Exam/.

---

Model development

Training

You can run training from the repository root with the CLI. All arguments are entirely optional; running the command without any flags will automatically train on the default mnist dataset:

# Run with absolute defaults (MNIST, 1 epoch, batch size 64, automatic device selection)
python -m src.training

# Run with custom configuration overrides
python -m src.training --dataset mnist --epochs 5 --batch-size 32 --device cuda

The checkpoint is written to weights/mnist.pth by default. The current training entry point supports mnist and can be extended with more datasets through the registry in src/training.py.

Testing

Run the basic tests with:

pytest -q

The tests are lightweight and only validate the training CLI, argument parsing, and factory wiring.

---

Contributing to the code

We use a branch → pull request → review workflow. All changes to main require at least one approved review — direct pushes are not allowed.

1. Open an issue describing your change
2. Create a branch and commit your work (reference the issue, e.g. fixes #5)
3. Open a pull request towards main
4. Get a review, address feedback, then merge

See CONTRIBUTION.md for the full guide.

---

Further use of the software

If you use this software in your research, teaching, or projects, please cite this repository. This project is released under the MIT License. You are free to use, modify, and distribute the software in accordance with the terms of the license.

Citation

APA:

Chen, S., Løkke, A., Gelato, R., Baburajan, R., Oei, K., & Catteau, M. (2026). Collaborative Coding Exam (Version 1.1.0) [Computer software]. https://github.com/FYS-8805-Collaborative-Coding/Collaborative-Coding-Exam

BibTex:

@software{Chen_Collaborative_Coding_Exam_2026,
author = {Chen, Siyan and Løkke, Andrea and Gelato, Riccardo and Baburajan, Rahul and Oei, Keyne and Catteau, Myrthe},
month = jun,
title = {{Collaborative Coding Exam}},
url = {https://github.com/FYS-8805-Collaborative-Coding/Collaborative-Coding-Exam},
version = {1.1.0},
year = {2026}
}

License

MIT