yaourt.tools.js_profile

Functions to process profiling output from onnxruntime.

yaourt.tools.js_profile.js_profile_to_dataframe(filename: str, as_df: bool = True, first_it_out: bool = False, agg: bool = False, agg_op_name: bool = False, with_shape: bool = False) → List | DataFrame

Reads an onnxruntime profiling file and returns its content.

Parameters:

• filename – filename holding the profiling stored in json format

• as_df – returns a DataFrame when True, otherwise a list of dicts

• first_it_out – if aggregated, leaves the first iteration out

• agg – aggregates by event

• agg_op_name – aggregates on operator name or operator index

• with_shape – keeps the shape before aggregating

Returns:

DataFrame or list of dicts

yaourt.tools.js_profile.plot_ort_profile(df: DataFrame, ax0: matplotlib.axes.Axes | None = None, ax1: matplotlib.axes.Axes | None = None, title: str | None = None) → matplotlib.axes.Axes

Plots time spent in computation based on a profiling dataframe.

The input dataframe should be produced by js_profile_to_dataframe().

Parameters:

• df – dataframe produced by js_profile_to_dataframe()

• ax0 – first axis to draw time (created if None)

• ax1 – second axis to draw occurrences (optional)

• title – graph title

Returns:

the axes used for drawing

yaourt.tools.js_profile.plot_ort_profile_timeline(df: DataFrame, ax: matplotlib.axes.Axes | None = None, iteration: int = -2, title: str | None = None, quantile: float = 0.5, fontsize: int = 12) → matplotlib.axes.Axes

Creates a timeline plot from an onnxruntime profiling dataframe.

The input dataframe should be produced by js_profile_to_dataframe().

Parameters:

• df – dataframe produced by js_profile_to_dataframe()

• ax – axis to draw (created if None)

• iteration – iteration to plot; negative values count from the end

• title – graph title

• quantile – operators below this quantile of duration are drawn in a different colour

• fontsize – font size

Returns:

the axis used for drawing

yaourt.tools.js_profile.post_process_df_profile(df: DataFrame, first_it_out: bool = False, agg: bool = False, agg_op_name: bool = True, with_shape: bool = False) → DataFrame

Post-processes a dataframe obtained after profiling onnxruntime.

Adds a column for a more explicit event name and adds a column for the iteration number.

Parameters:

• df – dataframe produced by js_profile_to_dataframe()

• agg – aggregates the result

• first_it_out – leaves the first iteration out of the aggregation

• agg_op_name – aggregates on operator name or operator index

• with_shape – keeps the shape to aggregate

Returns:

post-processed DataFrame