How faizanmasroor/handmade-neural-network works
This is an educational, from-scratch MNIST digit classifier. The documented workflow is to supply MNIST CSV files, load and preprocess the data, train or load a network, evaluate it, and optionally draw a digit for prediction. The sampled code supports these stages and the matrix-utility layer; several responsibilities share neural_network.py. The interactive entry-point code is only partly excerpted, so its broader wiring is not inferred.
Main entry points: neural_network.py Classifier runtime · matrix_math.py Matrix and loss math · data.py Data loader · pickles Prepared dataset
Components 10
MNIST workflow
- Classifier runtime
neural_network.py - Data loader
data.py - Prepared dataset
pickles - Digit drawing
neural_network.py
Network and math
- Network training
neural_network.py - Accuracy evaluation
neural_network.py - Digit prediction
neural_network.py - Matrix and loss math
matrix_math.py
Other components
- User
- MNIST CSV
Connections 13
- User → MNIST CSV: downloads
README.md - User → Classifier runtime: uses
README.md - MNIST CSV → Data loader: provides CSV
README.md - Classifier runtime → Data loader: loads data
neural_network.py - Data loader → Matrix and loss math: encodes labels
data.py - Classifier runtime → Prepared dataset: caches data
neural_network.py - Classifier runtime → Network training: trains or loads
neural_network.py - Network training → Matrix and loss math: uses activation and loss
neural_network.py - Classifier runtime → Accuracy evaluation: tests model
neural_network.py - Accuracy evaluation → Matrix and loss math: decodes predictions
neural_network.py - User → Digit drawing: draws digit
README.md - Digit drawing → Digit prediction: prepares input for
neural_network.py - Digit prediction → Matrix and loss math: decodes output
neural_network.py