.. DO NOT EDIT.
.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY.
.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE:
.. "auto_examples_patterns/plot_pattern_replay.py"
.. LINE NUMBERS ARE GIVEN BELOW.

.. only:: html

    .. note::
        :class: sphx-glr-download-link-note

        :ref:`Go to the end <sphx_glr_download_auto_examples_patterns_plot_pattern_replay.py>`
        to download the full example code.

.. rst-class:: sphx-glr-example-title

.. _sphx_glr_auto_examples_patterns_plot_pattern_replay.py:


.. _l-example-plot-pattern-replay:

Replaying graph-rewriting patterns
==================================

Every successful pattern optimization produces a
:class:`~onnx_light.onnx_core.optimization.LocalRewriting` record. These
records can reconstruct the optimized graph from the original model without
running pattern matching again.

.. GENERATED FROM PYTHON SOURCE LINES 12-21

.. code-block:: Python


    # sphinx_gallery_thumbnail_path = "_static/gallery_thumbnails/pattern_replay.png"

    from __future__ import annotations

    from onnx_light.onnx_lib import parser
    from onnx_light.onnx_core.optimization import GraphBuilder, GraphGraph, replay, standard_patterns
    from onnx_light.tools import pretty_onnx








.. GENERATED FROM PYTHON SOURCE LINES 22-27

Create and optimize the source model
++++++++++++++++++++++++++++++++++++

The source graph contains two type-preserving ``Cast`` nodes. The optimizer
rewrites them and returns the corresponding modification records.

.. GENERATED FROM PYTHON SOURCE LINES 27-47

.. code-block:: Python


    model = parser.parse_model(
        '<ir_version: 10, opset_import: ["" : 18]>\n'
        "agraph (float[4] x) => (float[4] y) {\n"
        "  casted = Cast <to=1> (x)\n"
        "  negated = Neg(casted)\n"
        "  y = Cast <to=1> (negated)\n"
        "}\n"
    )

    builder = GraphBuilder(model)
    graph = GraphGraph(builder, standard_patterns(["Cast"]))
    rewrites = graph.optimize()
    optimized_graph = builder.build_graph()

    print("Original graph:")
    print(pretty_onnx(model))
    print("Optimized graph:")
    print(pretty_onnx(builder.to_onnx("model")))





.. rst-class:: sphx-glr-script-out

 .. code-block:: none

    Original graph:
    opset: domain='' version=18
    graph: name='agraph'
    input: float[4] x
    0: Cast(x) -> casted
    1: Neg(casted) -> negated
    2: Cast(negated) -> y
    output: float[4] y
    Optimized graph:
    opset: domain='ai.onnx' version=18
    graph: name='agraph'
    input: float[4] x
    0: Neg(x) -> negated
    1: Identity(negated) -> y
    output: float[4] y




.. GENERATED FROM PYTHON SOURCE LINES 48-54

Inspect the captured modifications
++++++++++++++++++++++++++++++++++

Each record has a concise one-line display. Use ``to_detailed_string`` to
inspect the pattern, matched nodes, inserted nodes, and optimization iteration
needed to reproduce one modification.

.. GENERATED FROM PYTHON SOURCE LINES 54-59

.. code-block:: Python


    for rewrite in rewrites:
        print(rewrite)
        print(rewrite.to_detailed_string())





.. rst-class:: sphx-glr-script-out

 .. code-block:: none

    LocalRewriting(pattern=Cast, graph_path=<root>, matched_nodes=1, added_nodes=1)
    LocalRewriting:
      pattern: Cast
      graph_path: <root>
      iteration: 0
      matched_nodes:
        positions: [0]
      added_nodes:
        nodes: [Identity(outputs=[casted])]
        positions: [0]
      initializers:
        added: []
        added_positions: []
        removed: []
      value_renames: []
      timings:
        match_time_ns: 2695
        apply_time_ns: 3897
    LocalRewriting(pattern=Cast, graph_path=<root>, matched_nodes=1, added_nodes=1)
    LocalRewriting:
      pattern: Cast
      graph_path: <root>
      iteration: 0
      matched_nodes:
        positions: [2]
      added_nodes:
        nodes: [Identity(outputs=[y])]
        positions: [2]
      initializers:
        added: []
        added_positions: []
        removed: []
      value_renames: []
      timings:
        match_time_ns: 231
        apply_time_ns: 821
    LocalRewriting(pattern=RemoveIdentityNodes, graph_path=<root>, matched_nodes=3, added_nodes=2)
    LocalRewriting:
      pattern: RemoveIdentityNodes
      graph_path: <root>
      iteration: 1
      matched_nodes:
        positions: [0, 1, 2]
      added_nodes:
        nodes: [Neg(outputs=[negated]), Identity(outputs=[y])]
        positions: [0, 1]
      initializers:
        added: []
        added_positions: []
        removed: []
      value_renames: [casted->x]
      timings:
        match_time_ns: 0
        apply_time_ns: 0




.. GENERATED FROM PYTHON SOURCE LINES 60-66

Replay without matching patterns
++++++++++++++++++++++++++++++++

:func:`~onnx_light.onnx_core.optimization.replay` applies the captured
records to a fresh copy of the source model. The reconstructed graph is
byte-for-byte identical to the graph produced by the optimizer.

.. GENERATED FROM PYTHON SOURCE LINES 66-72

.. code-block:: Python


    replayed_graph = replay(model, rewrites)

    assert replayed_graph.SerializeToString() == optimized_graph.SerializeToString()
    print("Replayed graph:")
    print(pretty_onnx(replayed_graph))




.. rst-class:: sphx-glr-script-out

 .. code-block:: none

    Replayed graph:
    graph: name='agraph'
    input: float[4] x
    0: Neg(x) -> negated
    1: Identity(negated) -> y
    output: float[4] y





.. rst-class:: sphx-glr-timing

   **Total running time of the script:** (0 minutes 0.004 seconds)


.. _sphx_glr_download_auto_examples_patterns_plot_pattern_replay.py:

.. only:: html

  .. container:: sphx-glr-footer sphx-glr-footer-example

    .. container:: sphx-glr-download sphx-glr-download-jupyter

      :download:`Download Jupyter notebook: plot_pattern_replay.ipynb <plot_pattern_replay.ipynb>`

    .. container:: sphx-glr-download sphx-glr-download-python

      :download:`Download Python source code: plot_pattern_replay.py <plot_pattern_replay.py>`

    .. container:: sphx-glr-download sphx-glr-download-zip

      :download:`Download zipped: plot_pattern_replay.zip <plot_pattern_replay.zip>`


.. include:: plot_pattern_replay.recommendations


.. only:: html

 .. rst-class:: sphx-glr-signature

    `Gallery generated by Sphinx-Gallery <https://sphinx-gallery.github.io>`_