How antonistsiorvas/cifar10-classifiers works
This repository is CIFAR-10 coursework: it loads and prepares image data, compares classical 1-NN/3-NN and nearest-centroid baselines, and trains and evaluates configurable MLP/CNN models across experiments E1–E8. The neural workflow is experiment runner → training run → model and data preparation → saved metrics and predictions. The interim script separately evaluates raw and PCA-reduced data and can produce example predictions and per-class accuracy. Edges are limited to calls and data handling visible in excerpts or explicit README statements; assignment-2 and assignment-3 are marked TBA.
Main entry points: train.py Training and evaluation · knn_ncc.py NN and NCC · cifar10.py CIFAR loader · data.py Tensor preparation
Components 13
Dataset handling
- CIFAR loader
common/cifar10.py - Tensor preparation
assignment-1/data.py
Classical baselines
- NN and NCC
interim/knn_ncc.py - PCA analysis
interim/knn_ncc.py - Baseline reports
interim/knn_ncc.py
Neural experiments
- Experiment runner
assignment-1/experiments.py - Training and evaluation
assignment-1/train.py - MLP architecture
assignment-1/models.py - CNN architecture
assignment-1/models.py - Image augmentation
assignment-1/data.py - Run results
assignment-1/results
Other components
- User
- CIFAR-10 archive
Connections 13
- User → NN and NCC: runs baseline
README.md - User → Experiment runner: runs experiments
README.md - CIFAR-10 archive → CIFAR loader: downloads and extracts
common/cifar10.py - Tensor preparation → CIFAR loader: loads CIFAR-10
assignment-1/data.py - NN and NCC → CIFAR loader: loads CIFAR-10
interim/knn_ncc.py - NN and NCC → PCA analysis: optionally reduces
interim/knn_ncc.py - NN and NCC → Baseline reports: writes reports
interim/knn_ncc.py - Experiment runner → Training and evaluation: runs configurations
assignment-1/experiments.py - Training and evaluation → Tensor preparation: loads tensors
assignment-1/train.py - Training and evaluation → MLP architecture: constructs MLP
assignment-1/train.py - Training and evaluation → CNN architecture: constructs CNN
assignment-1/train.py - Training and evaluation → Image augmentation: augments batches
assignment-1/train.py - Training and evaluation → Run results: saves metrics and predictions
assignment-1/train.py