Skip to content

02-model-shapes.md

Latest commit

 

History

History
172 lines (142 loc) · 2.82 KB

02-model-shapes.md

File metadata and controls

172 lines (142 loc) · 2.82 KB
Raw

02 - Model Shapes

A "model shape" is a recurring way of composing the primitives. The same core runs all of them; the shape is a pattern, not a new abstraction.

1. Linear Transform Model

Input ───▶ Transform ───▶ Output

A single transform. See example 01.

2. Pipeline Model

Input
  ↓
Transform A
  ↓
State A
  ↓
Transform B
  ↓
State B
  ↓
Transform C
  ↓
Output

A linear chain. Each intermediate state is recorded in the trace. See example 02.

3. Evaluated Model

Input
  ↓
Transform Chain
  ↓
Output
  ↓
Evaluator
  ↓
EvaluationResult

See example 03.

4. Feedback Model

Input
  ↓
Transform Chain
  ↓
Output
  ↓
Evaluation
  ↓
Feedback
  ├── accept
  ├── retry
  ├── adjust
  ├── reject
  └── custom

See example 04.

5. Input Data Building Model

The most important shape. Documented in detail in 03-input-data-building-model.md.

Raw Input Data
  ↓
Normalize
  ↓
Data State 1
  ↓
Extract / Pick
  ↓
Data State 2
  ↓
Structure
  ↓
Built Data
  ↓
Evaluate
  ↓
Feedback

6. Mathematical Model

Input Vector
  ↓
Dense Layer
  ↓
Activation
  ↓
Hidden Representation
  ↓
Dense Layer
  ↓
Prediction
  ↓
Loss / Error

This is only one instance of the generic graph: the transforms happen to be dense layers and the evaluation happens to be a numeric loss. The library does not ship numeric layers — you implement transforms like these in a consumer.

7. Branching Model (future / advanced)

Input
  ├──▶ Transform B ──┐
  │                  ├──▶ Merge ──▶ Output
  └──▶ Transform C ──┘

Core v1 runs linear graphs only. Branching is documented as a future shape; it can be modeled today as a single branch+merge transform. See example 07.

8. Comparison Model

Input
  ├──▶ Graph A ──▶ Output A ──┐
  │                           ├──▶ Compare / Evaluate
  └──▶ Graph B ──▶ Output B ──┘

Run two separate graphs on the same input and compare. See example 08.

9. Lifecycle Model

Raw Run Data
  ↓
Extract Signals
  ↓
Measured State
  ↓
Evaluate
  ↓
Evaluation Result
  ↓
Resolve Feedback
  ↓
Next Action

Composable Model Graph does not define what a "run" is. It only provides the primitives to model a lifecycle.

Some systems do not improve by generating better outputs directly. They improve by making the lifecycle visible:

  • what happened?
  • what signals were measured?
  • what violated expectations?
  • what should update next?