Handwriting recognition (digits) classifier

What this project does

This small teaching project trains and evaluates an image classifier for handwritten characters using the scikit-learn digits dataset (8x8 images of handwritten digits 0–9). It demonstrates a reproducible training pipeline with a documented character vocabulary, a pseudo “cross-writer” split when writer identifiers are not available, and an editable suggestion-style output that provides top candidate characters and an “Unknown” fallback when the model is uncertain.

Data

The project uses the built-in sklearn.datasets.load_digits dataset (8x8 grayscale images of digits 0–9). The dataset does not include writer identifiers, so the code simulates writer groups deterministically from image statistics to demonstrate a group-based split that approximates cross-writer evaluation. This is a compromise for the classroom exercise; for real cross-writer validation use a dataset that contains explicit writer IDs.

How to build it

1. Create a Python 3.11+ virtual environment and activate it.
2. Install dependencies: pip install -r requirements.txt
3. Open project.py and study the run(seed: int = 42) function which builds a scaler + logistic-regression pipeline, and evaluates on a held-out group split.
4. Run the project with python project.py to see JSON metrics and example editable suggestions.

How to run

Run: python project.py The script prints a JSON object of metrics and a small preview of suggestions for held-out samples.

How to test

Run pytest in the project directory: pytest -q Tests assert presence of expected metrics, ranges, determinism, and a conservative accuracy floor.

Results

Measured by running python project.py (seed 42) on 2026-10-23; the tests in test_project.py assert the floor.

Metric	Value
dataset	sklearn_digits_8x8
n_train	1365
n_test	432
accuracy	0.9676
top3_accuracy	1
per_class_accuracy_mean	0.9693
unknown_fraction	0.0139

Also returned (see python project.py output): vocabulary (10 items), suggestions_preview (10 items).

Read the top3_accuracy value with care: the data here is small and synthetic with a planted, learnable structure, so a near-perfect score shows the pipeline works, not that the method would reach this on real data. See Limitations.

Limitations

• The dataset used is sklearn’s built-in digits collection (8x8 pixels) which is a small, pre-segmented digits dataset. It is not representative of real handwriting lines or multi-character transcription tasks.
• There are no true writer identifiers in the dataset; the project simulates writer groups from image statistics for pedagogical cross-writer splitting. This is explicitly noted and should be replaced with a dataset that contains writer IDs for rigorous cross-writer evaluation.
• The model is a compact classifier (logistic regression) aimed at reproducibility and speed for learning. It is not designed as production OCR and lacks image augmentation or a modern deep learning architecture.

Next steps

• Move to a dataset with explicit writer identifiers (for example IAM Handwriting Database) and run true cross-writer experiments.
• Replace the classifier with a CNN-based model (e.g. using PyTorch or TensorFlow) and add image augmentation to improve robustness.
• Extend the pipeline to return editable candidate suggestions integrated with a simple UI and incorporate language priors for real transcription tasks.

---

Full case study, build steps and evaluation checks: Handwriting recognition system on networkershome.com

Built by Networkers Home · Learn the skills in the AI Engineer Course · Part of 100 AI projects with code