# jkalogero/scalegmn architecture

> Architecture diagram of the GitHub repository jkalogero/scalegmn, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-06). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/jkalogero/scalegmn
- Repository: https://github.com/jkalogero/scalegmn

## Overview

ScaleGMN is a research codebase for learning over neural-network parameters, with experiments for INR classification, INR editing, and prediction of trained CNN generalization. The principal workflow is to configure an experiment, load parameter-space datasets, represent networks as graphs, and run ScaleGMN for task-specific training and evaluation; INR editing additionally applies predicted parameter updates to render images. Phase canonicalization is a separate documented preprocessing workflow. Edges are limited to excerpted imports/calls and README statements; unsampled implementation details are not inferred.

Main entry points: [`inr_editing.py`](https://github.com/jkalogero/scalegmn/blob/main/inr_editing.py) (INR editing), [`inr_classification.py`](https://github.com/jkalogero/scalegmn/blob/main/inr_classification.py) (INR classification), [`src/scalegmn/models.py`](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/models.py) (ScaleGMN models), [`predicting_generalization.py`](https://github.com/jkalogero/scalegmn/blob/main/predicting_generalization.py) (Generalization prediction)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_experiments["Experiment workflows"]
  node_configs["Experiment configs"]
  node_classification["INR classification"]
  node_editing["INR editing<br/>[inr_editing.py]"]
  node_generalization["Generalization prediction"]
end

subgraph group_data["Data processing"]
  node_canonicalization["Phase canonicalization"]
  node_phase_shift["Phase shift parameters<br/>[shift_params.py]"]
  node_dataset_factory["Dataset factory<br/>[__init__.py]"]
  node_inr_datasets["INR datasets"]
  node_cnn_datasets["CNN parameter datasets<br/>[cifar10_dataset.py]"]
  node_dataset_base["Dataset graph encoding<br/>[base_datasets.py]"]
  node_data_utils["Graph data utilities<br/>[data_utils.py]"]
end

subgraph group_model["ScaleGMN model"]
  node_scalegmn["ScaleGMN models<br/>[models.py]"]
  node_graph_init["Graph initialization<br/>[graph_init.py]"]
  node_gnn_layers["GNN message passing<br/>[gnn_layers.py]"]
  node_equivariant_layers["Equivariant layers<br/>[layers.py]"]
  node_mlp["MLP blocks<br/>[mlp.py]"]
  node_inr_runtime["SIREN runtime<br/>[inr.py]"]
end

subgraph group_runtime["Training components"]
  node_loss["Loss functions<br/>[loss.py]"]
  node_optimization["Optimizer setup<br/>[optim.py]"]
  node_cli["CLI arguments"]
  node_helpers["Experiment helpers<br/>[helpers.py]"]
end

node_researcher(("Experimenter"))

node_researcher -->|"selects"| node_configs
node_configs -->|"configures"| node_classification
node_configs -->|"configures"| node_editing
node_configs -->|"configures"| node_generalization
node_researcher -->|"runs preprocessing"| node_canonicalization
node_canonicalization -->|"uses"| node_phase_shift
node_canonicalization -->|"loads data"| node_dataset_factory
node_canonicalization -->|"uses"| node_loss
node_classification -->|"loads splits"| node_dataset_factory
node_editing -->|"loads splits"| node_dataset_factory
node_classification -->|"runs model"| node_scalegmn
node_editing -->|"runs model"| node_scalegmn
node_generalization -->|"uses model"| node_scalegmn
node_classification -->|"selects criterion"| node_loss
node_editing -->|"selects criterion"| node_loss
node_generalization -->|"uses loss"| node_loss
node_classification -->|"sets optimization"| node_optimization
node_editing -->|"sets optimization"| node_optimization
node_generalization -->|"sets optimization"| node_optimization
node_classification -->|"uses helpers"| node_helpers
node_editing -->|"uses helpers"| node_helpers
node_generalization -->|"uses helpers"| node_helpers
node_classification -->|"parses arguments"| node_cli
node_editing -->|"parses arguments"| node_cli
node_generalization -->|"parses arguments"| node_cli
node_dataset_factory -->|"dispatches"| node_inr_datasets
node_dataset_factory -->|"dispatches"| node_cnn_datasets
node_inr_datasets -->|"inherits"| node_dataset_base
node_cnn_datasets -->|"uses"| node_data_utils
node_dataset_base -->|"encodes graphs"| node_data_utils
node_scalegmn -->|"initializes graph"| node_graph_init
node_scalegmn -->|"constructs GNN"| node_gnn_layers
node_scalegmn -->|"constructs layers"| node_equivariant_layers
node_scalegmn -->|"uses graph utilities"| node_data_utils
node_gnn_layers -->|"uses blocks"| node_equivariant_layers
node_equivariant_layers -->|"builds MLPs"| node_mlp
node_equivariant_layers -->|"uses Sine"| node_inr_runtime
node_editing -->|"renders INRs"| node_inr_runtime
node_helpers -->|"uses argument helper"| node_cli

click node_configs "https://github.com/jkalogero/scalegmn/tree/main/configs"
click node_classification "https://github.com/jkalogero/scalegmn/blob/main/inr_classification.py"
click node_editing "https://github.com/jkalogero/scalegmn/blob/main/inr_editing.py"
click node_generalization "https://github.com/jkalogero/scalegmn/blob/main/predicting_generalization.py"
click node_canonicalization "https://github.com/jkalogero/scalegmn/blob/main/src/phase_canonicalization/canonicalization.py"
click node_phase_shift "https://github.com/jkalogero/scalegmn/blob/main/src/phase_canonicalization/shift_params.py"
click node_dataset_factory "https://github.com/jkalogero/scalegmn/blob/main/src/data/__init__.py"
click node_inr_datasets "https://github.com/jkalogero/scalegmn/tree/main/src/data"
click node_cnn_datasets "https://github.com/jkalogero/scalegmn/blob/main/src/data/cifar10_dataset.py"
click node_dataset_base "https://github.com/jkalogero/scalegmn/blob/main/src/data/base_datasets.py"
click node_data_utils "https://github.com/jkalogero/scalegmn/blob/main/src/data/data_utils.py"
click node_scalegmn "https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/models.py"
click node_graph_init "https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/graph_init.py"
click node_gnn_layers "https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/gnn_layers.py"
click node_equivariant_layers "https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/layers.py"
click node_mlp "https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/mlp.py"
click node_inr_runtime "https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/inr.py"
click node_loss "https://github.com/jkalogero/scalegmn/blob/main/src/utils/loss.py"
click node_optimization "https://github.com/jkalogero/scalegmn/blob/main/src/utils/optim.py"
click node_cli "https://github.com/jkalogero/scalegmn/blob/main/src/utils/setup_arg_parser.py"
click node_helpers "https://github.com/jkalogero/scalegmn/blob/main/src/utils/helpers.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_configs,node_classification,node_editing,node_generalization toneBlue
class node_canonicalization,node_phase_shift,node_dataset_factory,node_inr_datasets,node_cnn_datasets,node_dataset_base,node_data_utils toneAmber
class node_scalegmn,node_graph_init,node_gnn_layers,node_equivariant_layers,node_mlp,node_inr_runtime toneMint
class node_loss,node_optimization,node_cli,node_helpers toneRose
class node_researcher toneIndigo
```

## Components

### Experiment workflows

- [Experiment configs](https://github.com/jkalogero/scalegmn/tree/main/configs): `configs`
- [INR classification](https://github.com/jkalogero/scalegmn/blob/main/inr_classification.py): `inr_classification.py`
- [INR editing](https://github.com/jkalogero/scalegmn/blob/main/inr_editing.py): `inr_editing.py`
- [Generalization prediction](https://github.com/jkalogero/scalegmn/blob/main/predicting_generalization.py): `predicting_generalization.py`

### Data processing

- [Phase canonicalization](https://github.com/jkalogero/scalegmn/blob/main/src/phase_canonicalization/canonicalization.py): `src/phase_canonicalization/canonicalization.py`
- [Phase shift parameters](https://github.com/jkalogero/scalegmn/blob/main/src/phase_canonicalization/shift_params.py): `src/phase_canonicalization/shift_params.py`
- [Dataset factory](https://github.com/jkalogero/scalegmn/blob/main/src/data/__init__.py): `src/data/__init__.py`
- [INR datasets](https://github.com/jkalogero/scalegmn/tree/main/src/data): `src/data`
- [CNN parameter datasets](https://github.com/jkalogero/scalegmn/blob/main/src/data/cifar10_dataset.py): `src/data/cifar10_dataset.py`
- [Dataset graph encoding](https://github.com/jkalogero/scalegmn/blob/main/src/data/base_datasets.py): `src/data/base_datasets.py`
- [Graph data utilities](https://github.com/jkalogero/scalegmn/blob/main/src/data/data_utils.py): `src/data/data_utils.py`

### ScaleGMN model

- [ScaleGMN models](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/models.py): `src/scalegmn/models.py`
- [Graph initialization](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/graph_init.py): `src/scalegmn/graph_init.py`
- [GNN message passing](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/gnn_layers.py): `src/scalegmn/gnn_layers.py`
- [Equivariant layers](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/layers.py): `src/scalegmn/layers.py`
- [MLP blocks](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/mlp.py): `src/scalegmn/mlp.py`
- [SIREN runtime](https://github.com/jkalogero/scalegmn/blob/main/src/scalegmn/inr.py): `src/scalegmn/inr.py`

### Training components

- [Loss functions](https://github.com/jkalogero/scalegmn/blob/main/src/utils/loss.py): `src/utils/loss.py`
- [Optimizer setup](https://github.com/jkalogero/scalegmn/blob/main/src/utils/optim.py): `src/utils/optim.py`
- [CLI arguments](https://github.com/jkalogero/scalegmn/blob/main/src/utils/setup_arg_parser.py): `src/utils/setup_arg_parser.py`
- [Experiment helpers](https://github.com/jkalogero/scalegmn/blob/main/src/utils/helpers.py): `src/utils/helpers.py`

### Other components

- **Experimenter**

## Connections

- Experimenter → Experiment configs: selects (evidence: `README.md`)
- Experiment configs → INR classification: configures (evidence: `README.md`)
- Experiment configs → INR editing: configures (evidence: `README.md`)
- Experiment configs → Generalization prediction: configures (evidence: `README.md`)
- Experimenter → Phase canonicalization: runs preprocessing (evidence: `README.md`)
- Phase canonicalization → Phase shift parameters: uses (evidence: `src/phase_canonicalization/canonicalization.py`)
- Phase canonicalization → Dataset factory: loads data (evidence: `src/phase_canonicalization/canonicalization.py`)
- Phase canonicalization → Loss functions: uses (evidence: `src/phase_canonicalization/canonicalization.py`)
- INR classification → Dataset factory: loads splits (evidence: `inr_classification.py`)
- INR editing → Dataset factory: loads splits (evidence: `inr_editing.py`)
- INR classification → ScaleGMN models: runs model (evidence: `inr_classification.py`)
- INR editing → ScaleGMN models: runs model (evidence: `inr_editing.py`)
- Generalization prediction → ScaleGMN models: uses model (evidence: `predicting_generalization.py`)
- INR classification → Loss functions: selects criterion (evidence: `inr_classification.py`)
- INR editing → Loss functions: selects criterion (evidence: `inr_editing.py`)
- Generalization prediction → Loss functions: uses loss (evidence: `predicting_generalization.py`)
- INR classification → Optimizer setup: sets optimization (evidence: `inr_classification.py`)
- INR editing → Optimizer setup: sets optimization (evidence: `inr_editing.py`)
- Generalization prediction → Optimizer setup: sets optimization (evidence: `predicting_generalization.py`)
- INR classification → Experiment helpers: uses helpers (evidence: `inr_classification.py`)
- INR editing → Experiment helpers: uses helpers (evidence: `inr_editing.py`)
- Generalization prediction → Experiment helpers: uses helpers (evidence: `predicting_generalization.py`)
- INR classification → CLI arguments: parses arguments (evidence: `inr_classification.py`)
- INR editing → CLI arguments: parses arguments (evidence: `inr_editing.py`)
- Generalization prediction → CLI arguments: parses arguments (evidence: `predicting_generalization.py`)
- Dataset factory → INR datasets: dispatches (evidence: `src/data/__init__.py`)
- Dataset factory → CNN parameter datasets: dispatches (evidence: `src/data/__init__.py`)
- INR datasets → Dataset graph encoding: inherits (evidence: `src/data/mnist_inr_dataset.py`)
- CNN parameter datasets → Graph data utilities: uses (evidence: `src/data/cifar10_dataset.py`)
- Dataset graph encoding → Graph data utilities: encodes graphs (evidence: `src/data/base_datasets.py`)
- ScaleGMN models → Graph initialization: initializes graph (evidence: `src/scalegmn/models.py`)
- ScaleGMN models → GNN message passing: constructs GNN (evidence: `src/scalegmn/models.py`)
- ScaleGMN models → Equivariant layers: constructs layers (evidence: `src/scalegmn/models.py`)
- ScaleGMN models → Graph data utilities: uses graph utilities (evidence: `src/scalegmn/models.py`)
- GNN message passing → Equivariant layers: uses blocks (evidence: `src/scalegmn/gnn_layers.py`)
- Equivariant layers → MLP blocks: builds MLPs (evidence: `src/scalegmn/layers.py`)
- Equivariant layers → SIREN runtime: uses Sine (evidence: `src/scalegmn/layers.py`)
- INR editing → SIREN runtime: renders INRs (evidence: `inr_editing.py`)
- Experiment helpers → CLI arguments: uses argument helper (evidence: `src/utils/helpers.py`)
