# unbailanlinch/21days architecture

> Architecture diagram of the GitHub repository unbailanlinch/21days, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-09-30). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/unbailanlinch/21days
- Repository: https://github.com/unbailanlinch/21days

## Overview

This repository is a Unity game project, but the sampled runtime sources describe a substantial companion subsystem: expression-recognition model development for character face rigs. Its workflow builds canonical expression data and rig projections, trains and calibrates a classifier, evaluates it, and exports a model for deployment. The graph focuses on that evidenced subsystem rather than inferring Unity gameplay connections from the tree. Edges are limited to relationships documented by source excerpts or the SOURCE INDEX; other Unity and harness capabilities are not mapped because their runtime sources were not sampled.

Main entry points: [`ML/expression-recognition/exprnet/train.py`](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/train.py) (Training pipeline), [`ML/expression-recognition/exprnet/evaluate.py`](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/evaluate.py) (Model evaluation), [`ML/expression-recognition/exprnet/datasets.py`](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/datasets.py) (Dataset loading), [`ML/expression-recognition/analysis/laila_upper_bound/run.py`](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/analysis/laila_upper_bound/run.py) (Rig upper-bound analysis)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_representation["Expression representation"]
  node_canonical["Canonical space<br/>[canonical.py]"]
  node_rig["Face rig<br/>[rig.py]"]
  node_common["Labels and config<br/>[common.py]"]
end

subgraph group_data["Data preparation"]
  node_datasets["Dataset loading<br/>[datasets.py]"]
  node_extract["Feature extraction<br/>[extract.py]"]
  node_facs["FACS synthesis<br/>[facs.py]"]
  node_augment["Data augmentation<br/>[augment.py]"]
  node_coverage["Rig coverage<br/>[coverage.py]"]
end

subgraph group_learning["Model learning"]
  node_models["Classifier models<br/>[models.py]"]
  node_losses["Training losses<br/>[losses.py]"]
  node_train["Training pipeline<br/>[train.py]"]
end

subgraph group_validation["Validation and export"]
  node_calibrate["Calibration and rejection<br/>[calibrate.py]"]
  node_golden["Golden-set checks<br/>[golden_check.py]"]
  node_evaluate["Model evaluation<br/>[evaluate.py]"]
  node_export["Model export<br/>[export.py]"]
  node_analysis["Rig upper-bound analysis<br/>[run.py]"]
end

node_researcher(("ML engineer"))

node_researcher -.->|"starts training"| node_train
node_train -->|"loads labels"| node_common
node_train -->|"loads dimensions"| node_canonical
node_train -->|"loads rig"| node_rig
node_train -->|"synthesizes samples"| node_facs
node_train -->|"builds training data"| node_datasets
node_train -->|"augments batches"| node_augment
node_train -->|"trains classifier"| node_models
node_train -->|"computes loss"| node_losses
node_train -->|"fits thresholds"| node_calibrate
node_train -->|"evaluates run"| node_evaluate
node_datasets -->|"aligns features"| node_canonical
node_datasets -->|"projects features"| node_rig
node_facs -->|"uses expression space"| node_canonical
node_coverage -->|"projects samples"| node_rig
node_coverage -->|"reads cached features"| node_datasets
node_coverage -->|"generates prototypes"| node_facs
node_evaluate -->|"applies thresholds"| node_calibrate
node_evaluate -->|"checks golden set"| node_golden
node_evaluate -->|"loads classifier"| node_models
node_evaluate -->|"compares deployment graph"| node_export
node_export -->|"exports classifier"| node_models
node_analysis -->|"runs training"| node_train
node_analysis -->|"rebuilds splits"| node_datasets
node_analysis -->|"projects inputs"| node_rig
node_analysis -->|"compares predictions"| node_evaluate

click node_canonical "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/canonical.py"
click node_rig "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/rig.py"
click node_common "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/common.py"
click node_datasets "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/datasets.py"
click node_extract "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/extract.py"
click node_facs "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/facs.py"
click node_augment "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/augment.py"
click node_coverage "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/coverage.py"
click node_models "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/models.py"
click node_losses "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/losses.py"
click node_train "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/train.py"
click node_calibrate "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/calibrate.py"
click node_golden "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/golden_check.py"
click node_evaluate "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/evaluate.py"
click node_export "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/export.py"
click node_analysis "https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/analysis/laila_upper_bound/run.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_canonical,node_rig,node_common toneBlue
class node_datasets,node_extract,node_facs,node_augment,node_coverage toneAmber
class node_models,node_losses,node_train toneMint
class node_calibrate,node_golden,node_evaluate,node_export,node_analysis toneRose
class node_researcher toneIndigo
```

## Components

### Expression representation

- [Canonical space](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/canonical.py): `ML/expression-recognition/exprnet/canonical.py`
- [Face rig](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/rig.py): `ML/expression-recognition/exprnet/rig.py`
- [Labels and config](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/common.py): `ML/expression-recognition/exprnet/common.py`

### Data preparation

- [Dataset loading](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/datasets.py): `ML/expression-recognition/exprnet/datasets.py`
- [Feature extraction](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/extract.py): `ML/expression-recognition/exprnet/extract.py`
- [FACS synthesis](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/facs.py): `ML/expression-recognition/exprnet/facs.py`
- [Data augmentation](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/augment.py): `ML/expression-recognition/exprnet/augment.py`
- [Rig coverage](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/coverage.py): `ML/expression-recognition/exprnet/coverage.py`

### Model learning

- [Classifier models](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/models.py): `ML/expression-recognition/exprnet/models.py`
- [Training losses](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/losses.py): `ML/expression-recognition/exprnet/losses.py`
- [Training pipeline](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/train.py): `ML/expression-recognition/exprnet/train.py`

### Validation and export

- [Calibration and rejection](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/calibrate.py): `ML/expression-recognition/exprnet/calibrate.py`
- [Golden-set checks](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/golden_check.py): `ML/expression-recognition/exprnet/golden_check.py`
- [Model evaluation](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/evaluate.py): `ML/expression-recognition/exprnet/evaluate.py`
- [Model export](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/exprnet/export.py): `ML/expression-recognition/exprnet/export.py`
- [Rig upper-bound analysis](https://github.com/unbailanlinch/21days/blob/main/ML/expression-recognition/analysis/laila_upper_bound/run.py): `ML/expression-recognition/analysis/laila_upper_bound/run.py`

### Other components

- **ML engineer**

## Connections

- ML engineer → Training pipeline: starts training
- Training pipeline → Labels and config: loads labels (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Canonical space: loads dimensions (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Face rig: loads rig (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → FACS synthesis: synthesizes samples (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Dataset loading: builds training data (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Data augmentation: augments batches (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Classifier models: trains classifier (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Training losses: computes loss (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Calibration and rejection: fits thresholds (evidence: `ML/expression-recognition/exprnet/train.py`)
- Training pipeline → Model evaluation: evaluates run (evidence: `ML/expression-recognition/exprnet/train.py`)
- Dataset loading → Canonical space: aligns features (evidence: `ML/expression-recognition/exprnet/datasets.py`)
- Dataset loading → Face rig: projects features (evidence: `ML/expression-recognition/exprnet/datasets.py`)
- FACS synthesis → Canonical space: uses expression space (evidence: `ML/expression-recognition/exprnet/facs.py`)
- Rig coverage → Face rig: projects samples (evidence: `ML/expression-recognition/exprnet/coverage.py`)
- Rig coverage → Dataset loading: reads cached features (evidence: `ML/expression-recognition/exprnet/coverage.py`)
- Rig coverage → FACS synthesis: generates prototypes (evidence: `ML/expression-recognition/exprnet/coverage.py`)
- Model evaluation → Calibration and rejection: applies thresholds (evidence: `ML/expression-recognition/exprnet/evaluate.py`)
- Model evaluation → Golden-set checks: checks golden set (evidence: `ML/expression-recognition/exprnet/evaluate.py`)
- Model evaluation → Classifier models: loads classifier (evidence: `ML/expression-recognition/exprnet/evaluate.py`)
- Model evaluation → Model export: compares deployment graph (evidence: `ML/expression-recognition/exprnet/evaluate.py`)
- Model export → Classifier models: exports classifier (evidence: `ML/expression-recognition/exprnet/export.py`)
- Rig upper-bound analysis → Training pipeline: runs training (evidence: `ML/expression-recognition/analysis/laila_upper_bound/run.py`)
- Rig upper-bound analysis → Dataset loading: rebuilds splits (evidence: `ML/expression-recognition/analysis/laila_upper_bound/run.py`)
- Rig upper-bound analysis → Face rig: projects inputs (evidence: `ML/expression-recognition/analysis/laila_upper_bound/run.py`)
- Rig upper-bound analysis → Model evaluation: compares predictions (evidence: `ML/expression-recognition/analysis/laila_upper_bound/run.py`)
