ml_switcheroo.plugins.keras_sequential

Plugin for Keras Sequential Container Translation.

This plugin bridges the difference between PyTorch’s nn.Sequential(*layers) (variadic args) and Keras’s keras.Sequential([layers]) (list input).

It performs two key transformations:

  1. API Renaming: Swaps the function name to keras.Sequential (or configured API).

  2. Argument Packing: Collects all positional arguments (individual layers) into a single list argument to match the Keras constructor signature.

Functions

transform_keras_sequential(→ libcst.Call)

Plugin Hook: Transforms Sequential container initialization.

Module Contents

ml_switcheroo.plugins.keras_sequential.transform_keras_sequential(node: libcst.Call, ctx: ml_switcheroo.core.hooks.HookContext) → libcst.Call[source]

Plugin Hook: Transforms Sequential container initialization.

Transformation:

Input: Sequential(layer1, layer2, …) Output: keras.Sequential([layer1, layer2, …])

This hook is triggered for operations mapped with requires_plugin=”keras_sequential_pack”. It uses ctx.lookup_api(“Sequential”) to determine the target class name, defaulting to keras.Sequential if lookup fails.

Parameters:

  • node – The original function call node.

  • ctx – The plugin execution context.

Returns:

The transformed function call.