How wiegerw/nerva works
This repository contributes dataset preparation utilities to the broader Nerva neural-network project; it does not contain the neural-network libraries themselves. The principal workflow is selecting MNIST or CIFAR-10 through the CLI, optionally downloading source data, then producing flattened, normalized NPZ datasets. MNIST is loaded directly from its archive; CIFAR-10 is extracted and its batches are assembled. The graph covers only the sampled preparation script, not downstream training consumers.
Main entry points: prepare_data.py Dataset CLI
Components 12
Other components
- CLI user
- Dataset CLI
data/prepare_data.py - MNIST download
data/prepare_data.py - MNIST archive
- MNIST conversion
data/prepare_data.py - CIFAR download
data/prepare_data.py - CIFAR archive
- CIFAR conversion
data/prepare_data.py - Batch loader
data/prepare_data.py - Prepared NPZ
- MNIST host
- CIFAR host
Connections 14
- CLI user → Dataset CLI: selects dataset
data/prepare_data.py - Dataset CLI → MNIST download: optionally calls
data/prepare_data.py - Dataset CLI → MNIST conversion: calls
data/prepare_data.py - MNIST download → MNIST host: downloads from
data/prepare_data.py - MNIST download → MNIST archive: writes archive
data/prepare_data.py - MNIST conversion → MNIST archive: reads archive
data/prepare_data.py - MNIST conversion → Prepared NPZ: writes NPZ
data/prepare_data.py - Dataset CLI → CIFAR download: optionally calls
data/prepare_data.py - Dataset CLI → CIFAR conversion: calls
data/prepare_data.py - CIFAR download → CIFAR host: downloads from
data/prepare_data.py - CIFAR download → CIFAR archive: stores and extracts
data/prepare_data.py - CIFAR conversion → Batch loader: calls
data/prepare_data.py - Batch loader → CIFAR archive: reads batches
data/prepare_data.py - CIFAR conversion → Prepared NPZ: writes NPZ
data/prepare_data.py