How aryan-sz13/aiml-recruitment-2026-aryan-singh works
This repository is a Python/TensorFlow MNIST digit-classification and analysis application. Its main workflow loads and preprocesses MNIST, runs a controlled model-ablation suite, and uses the results for metric analysis and diagnostic plots. The README also documents dataset exploration, training/validation curves, confusion matrices, per-class metrics, and error inspection. Source excerpts show the main orchestration and data-loading calls; the experiment module’s indexed references support its core dependencies, but not every internal call detail is shown.
Main entry points: experiment.py Ablation suite · main.py Pipeline entry · data.py Data preparation · mnist_download.py Dataset download
Components 12
Data ingestion
- Dataset download
src/mnist_download.py - IDX parser
src/mnist_parser.py - Data preparation
src/data.py
Modeling and experiments
- MLP builder
src/model.py - Model training
src/train.py - Ablation suite
src/experiment.py
Evaluation and visualization
- Metric analysis
src/evaluate.py - Diagnostic plots
src/visualize.py
Application runtime
- Run configuration
src/config.py - Pipeline entry
main.py
Other components
- Assessment user
- MNIST mirror
Connections 14
- Assessment user → Pipeline entry: runs pipeline
main.py - Pipeline entry → Run configuration: ensures directories
main.py - Pipeline entry → Data preparation: loads data
main.py - Data preparation → Dataset download: downloads missing files
src/data.py - MNIST mirror → Dataset download: provides archives
src/mnist_download.py - Data preparation → IDX parser: loads IDX samples
src/data.py - Pipeline entry → Ablation suite: runs suite
main.py - Ablation suite → Run configuration: uses settings
src/experiment.py - Ablation suite → Data preparation: loads data
src/experiment.py - Ablation suite → MLP builder: builds variants
src/experiment.py - Ablation suite → Model training: trains variants
src/experiment.py - Ablation suite → Metric analysis: evaluates models
src/experiment.py - Pipeline entry → Metric analysis: finds error cases
main.py - Pipeline entry → Diagnostic plots: renders diagnostics
main.py