How zhiyi1214/neuronetwork works
NeuroStudio is a browser-based MNIST experimentation platform with two main user workflows: compare saved models on hand-drawn digits, and assemble/train networks while observing metrics and learned features. The graph includes the documented MLP, CNN, and autoencoder training paths, plus offline evaluation. Source excerpts show overlapping WebUI backends; exact frontend-to-backend wiring and some training internals are not fully visible, so uncertain links are omitted.
Main entry points: weights Saved models · app.py Studio API · index.html Studio interface · webui_server.py WebUI service
Components 17
Browser studio
- Studio interface
templates/index.html - Digit pixels
templates/index.html
Inference and models
- WebUI service
webui_server.py - Studio API
backend/app.py - Model inference
webui_server.py - Feature maps
backend/app.py
Training studio
- Training worker
backend/trainer.py - Architecture compiler
backend/compiler.py - MLP trainer
train.py - CNN trainer
train_cnn.py - Offline evaluation
evaluate.py
Datasets and artifacts
- Saved models
weights - Run metadata
weights/manifest.json - MNIST loading
backend/datasets.py - Training images
train - Test images
test
Other components
- Experimenter
Connections 23
- Experimenter → Studio interface: uses
- Studio interface → Digit pixels: edits pixels
- Studio interface → Studio API: sends actions
- Studio interface → WebUI service: requests inference
- Studio API → Training worker: starts and controls
- Studio API → Feature maps: serves maps
- Studio API → Saved models: lists and deletes
- Studio API → Run metadata: updates records
- WebUI service → Model inference: dispatches prediction
- WebUI service → Saved models: loads models
- WebUI service → Run metadata: reads metadata
- Model inference → Saved models: loads parameters
- Training worker → Architecture compiler: builds networks
- Training worker → MNIST loading: requests batches
- MNIST loading → Training images: reads
- MNIST loading → Test images: reads
- MLP trainer → Training images: reads
- MLP trainer → Saved models: saves model
- CNN trainer → Training images: reads
- CNN trainer → Saved models: saves model
- Offline evaluation → Test images: reads
- Offline evaluation → Saved models: loads model
- Offline evaluation → Model inference: evaluates batches