Application layout
SignalWeave Modern is organized into three zones:
- Top toolbar: project-level actions
- Middle workspace: left inspection area and right control stack
- Bottom output area: console and error graph
The left workspace is used for visualization and analysis. The right stack is used for configuration, training, testing, pattern editing, and summary. The bottom is used for logs and error progression.
Top toolbar
- ✚ New: create a fresh project
- 📂 Load: load an existing project file
- 💾 Save: save the current project
- 🖫 Save As: save to a new path
- Status text: shows whether the workspace is ready, unsaved, or loaded from a file
The toolbar operates on the entire project, not on separate network, weight, or pattern files.
Network tab
The Network tab defines the model itself.
- Project name: the display name stored in the project
- Network type:
FeedForwardorSimpleRecurrent - Inputs: number of input units
- Hidden layer 1: size of the first hidden layer
- Hidden layer 2: size of the second hidden layer when allowed
- Outputs: number of output units
- Biases: input, hidden 1, and hidden 2 bias toggles
- Weight range: random range used when weights are initialized or reset
Changing these settings changes the topology, graph layout, weight matrix shapes, and pattern validation rules.
Training tab
The Training tab controls how the model learns.
- Learning rate: scale of each weight update
- Momentum: carry-over from previous updates
- Error threshold: target error level
- Learning steps: length of one training block
- Batch update: switch between batch and per-pattern updates
- Cross entropy: switch loss function
- Reset: rebuild weights from fresh random values
- Train #N: continue training from current weights
The tab also contains the progress readout, progress bar, training-session list, and rollback button.
Tests tab
The Tests tab is the pattern-by-pattern inspection workflow.
- Test selected: run the currently selected pattern
- Test all: run the full dataset
- Pattern list: shows each example and its latest status
- Pattern inspector: shows inputs, targets, outputs, delta, hidden activations, and current error
Selecting a pattern also updates the main network graph with that pattern’s latest cached result.
Patterns tab
The Patterns tab edits the dataset stored inside the project.
- Text mode: raw text editing
- Graphic mode: row-and-cell editing in a table view
- Add pattern: append a new pattern row
- Graphic row controls: label editor, reset checkbox, binary input toggles, binary target toggles, delete action
Both modes edit the same live pattern set and stay synchronized.
Weights workspace
The Weights workspace is the dedicated matrix inspector.
- Source: current weights or a saved training checkpoint
- Layer: choose which matrix to inspect
- View:
Heatmap,Magnitude, orValues - Stats panel: matrix dimensions, min, max, mean absolute weight, near-zero count
Use this when the main graph is too dense for detailed weight inspection.
Analysis workspace
The Analysis workspace is for whole-dataset inspection.
- Output clustering
- Hidden-state clustering
- Compatibility summary
- Time series
- Surface plot
Most analysis modes can run against either the current weights or an earlier Train #N checkpoint.
Exports
- Network graph: PNG
- Error graph: PNG
- Console: Markdown file
- Analysis: text, CSV, or PNG depending on mode
- Hidden activations: selected pattern or full set as CSV
Complete reference
This site page is the public web manual. The repository also contains the full control-by-control reference used by the project itself:
Open the repository manual source
That reference covers every visible control in the current Modern UI, including tabs, splitters, pattern-table controls, export buttons, and status displays.