ml_switcheroo.core.latex

LaTeX DSL Core Package.

This package contains the semantic primitives for the Machine Intelligence Definition Language (MIDL). It defines the node structures mirroring the LaTeX macros (Attribute, Op, etc.) used to visualize neural network architectures.

Submodules

• ml_switcheroo.core.latex.nodes
• ml_switcheroo.core.latex.parser

Classes

LatexNode	Abstract base class for all MIDL nodes.
ModelContainer	Root container representing the Model definition block.
MemoryNode	Represents stateful memory allocation (e.g., Weights/Layers).
InputNode	Represents the model input definition.
ComputeNode	Represents a stateless operation call.
StateOpNode	Represents a call to a stateful layer defined in Memory.
ReturnNode	Represents the output return statement.

Package Contents

class ml_switcheroo.core.latex.LatexNode

Bases: abc.ABC

Abstract base class for all MIDL nodes.

Enforces a to_latex() method for serialization support.

abstractmethod to_latex() → str

Serializes the node object back into its LaTeX macro representation.

Returns:

Valid LaTeX code string.

Return type:

str

class ml_switcheroo.core.latex.ModelContainer

Bases: LatexNode

Root container representing the Model definition block. Maps to the DefModel environment.

name: str

The model class name.

children: List[LatexNode] = []

List of body statements (Memory, Input, Ops, Return).

to_latex() → str

Render the full \begin{DefModel}...\end{DefModel} block.

class ml_switcheroo.core.latex.MemoryNode

Bases: LatexNode

Represents stateful memory allocation (e.g., Weights/Layers). Maps to the \Attribute macro.

Example:

\Attribute{conv}{Conv2d}{in=1, out=32, k=3}

node_id: str

The unique identifier for the attribute.

op_type: str

The operation type (e.g., ‘Conv2d’).

config: Dict[str, str]

Configuration parameters for the layer.

to_latex() → str

Render to \Attribute macro.

class ml_switcheroo.core.latex.InputNode

Bases: LatexNode

Represents the model input definition. Maps to the \Input macro.

Example:

\Input{x}{[B, 1, 28, 28]}

name: str

Name of the input variable.

shape: str

Shape descriptor string.

to_latex() → str

Render to \Input macro.

class ml_switcheroo.core.latex.ComputeNode

Bases: LatexNode

Represents a stateless operation call. Maps to the \Op macro.

Example:

\Op{s2}{Flatten}{s1, start=1}{[B, 21632]}

node_id: str

The unique identifier to assign the result to.

op_type: str

The operation type (e.g., ‘Flatten’).

args: List[str]

List of arguments passed to the operation.

shape: str

Resulting shape descriptor.

to_latex() → str

Render to \Op macro.

class ml_switcheroo.core.latex.StateOpNode

Bases: LatexNode

Represents a call to a stateful layer defined in Memory. Maps to the \StateOp macro.

Example:

\StateOp{s1}{conv}{x}{[B, 32, 26, 26]}

node_id: str

The unique identifier to assign the result to.

attribute_id: str

The ID of the attribute being called.

args: List[str]

List of arguments passed to the call.

shape: str

Resulting shape descriptor.

to_latex() → str

Render to \StateOp macro.

class ml_switcheroo.core.latex.ReturnNode

Bases: LatexNode

Represents the output return statement. Maps to the \Return macro.

Example:

\Return{s3}

target_id: str

The variable ID to return.

to_latex() → str

Render to \Return macro.