How dacorvo/tensorsandbox works
This repository is a TensorFlow CIFAR-10 image-classification project. Its main workflow builds input batches, selects a CNN, trains it and saves checkpoints; evaluation restores checkpoints and reports predictions and accuracy. It also contains several selectable CNN families, graph-freezing and model-inspection utilities, and distributed-training support. The graph focuses on the CIFAR-10 runtime; source excerpts do not show all utility or model-family internals, so uncertain links are omitted.
Main entry points: select.py Model selector · model.py CNN base layers · train.py Training loop · eval.py Evaluation loop
Components 19
Data input
- Input pipeline
cifar10/models/data.py
Model architectures
- Model selector
cifar10/models/select.py - CNN base layers
cifar10/models/model.py - AlexNet variants
cifar10/models/alex - Network-in-Network variants
cifar10/models/nin - SqueezeNet variants
cifar10/models/squeeze - ConvOne
cifar10/models/convone.py - CS231n CNN
cifar10/models/cs231n.py - Tutorial CNN
cifar10/models/tuto.py
Training and evaluation
- Training loop
cifar10/train.py - Cluster training
cifar10/train_cluster.py - Run configuration
cifar10/job.py - Model checkpoints
cifar10 - Evaluation loop
cifar10/eval.py
Model utilities
- Graph freezing
cifar10/freeze_graph.py - Model info
cifar10/info.py
Other components
- Operator
- CIFAR-10 files
- Dataset download
Connections 33
- Operator → Training loop: starts training
cifar10/train.py - Operator → Evaluation loop: starts evaluation
cifar10/eval.py - Training loop → Input pipeline: requests batches
cifar10/train.py - Evaluation loop → Input pipeline: requests eval data
cifar10/eval.py - Input pipeline → CIFAR-10 files: reads records
cifar10/models/data.py - Input pipeline → Dataset download: downloads archive
cifar10/models/data.py - Training loop → Model selector: selects model
cifar10/train.py - Evaluation loop → Model selector: selects model
cifar10/eval.py - Model selector → AlexNet variants: dispatches variants
cifar10/models/select.py - Model selector → Network-in-Network variants: dispatches variants
cifar10/models/select.py - Model selector → SqueezeNet variants: dispatches variants
cifar10/models/select.py - Model selector → ConvOne: selects model
cifar10/models/select.py - Model selector → CS231n CNN: selects model
cifar10/models/select.py - Model selector → Tutorial CNN: selects model
cifar10/models/select.py - Training loop → CNN base layers: builds inference and loss
cifar10/train.py - Evaluation loop → CNN base layers: builds inference
cifar10/eval.py - AlexNet variants → CNN base layers: inherits layers
cifar10/models/alex/alex1.py - Network-in-Network variants → CNN base layers: inherits layers
cifar10/models/nin/nin.py - SqueezeNet variants → CNN base layers: inherits layers
cifar10/models/squeeze/squeezenet.py - ConvOne → CNN base layers: inherits layers
cifar10/models/convone.py - CS231n CNN → CNN base layers: inherits layers
cifar10/models/cs231n.py - Tutorial CNN → CNN base layers: inherits layers
cifar10/models/tuto.py - Training loop → Model checkpoints: saves checkpoints
cifar10/train.py - Evaluation loop → Model checkpoints: restores checkpoint
cifar10/eval.py - Cluster training → Training loop: delegates training
cifar10/train_cluster.py - Training loop → Run configuration: imports config
cifar10/train.py - Evaluation loop → Run configuration: uses log config
cifar10/eval.py - Operator → Graph freezing: freezes graph
cifar10/freeze_graph.py - Graph freezing → Model selector: selects model
cifar10/freeze_graph.py - Graph freezing → Run configuration: uses run config
cifar10/freeze_graph.py - Operator → Model info: inspects model
cifar10/info.py - Model info → Model selector: selects model
cifar10/info.py - Model info → Input pipeline: uses data utilities
cifar10/info.py