module testing.einsum.einsum_bench#

Short summary#

module mlprodict.testing.einsum.einsum_bench

Function to measure the performance of einsum decomposition.

source on GitHub

Functions#

function

truncated documentation

_make_einsum_model

_make_inputs

_measure_time

Measures a statement and returns the results as a dictionary.

einsum_benchmark

Investigates whether or not the decomposing einsum is faster.

Documentation#

Function to measure the performance of einsum decomposition.

source on GitHub

mlprodict.testing.einsum.einsum_bench._make_einsum_model(equation, opset=17)#
mlprodict.testing.einsum.einsum_bench._make_inputs(equation, shapes)#
mlprodict.testing.einsum.einsum_bench._measure_time(stmt, *x, repeat=5, number=5, div_by_number=True, first_run=True, max_time=None)#

Measures a statement and returns the results as a dictionary.

Parameters:

  • stmt – string

  • *x –

    inputs

  • repeat – average over repeat experiment

  • number – number of executions in one row

  • div_by_number – divide by the number of executions

  • first_run – if True, runs the function once before measuring

  • max_time – execute the statement until the total goes beyond this time (approximatively), repeat is ignored, div_by_number must be set to True

Returns:

dictionary

See Timer.repeat for a better understanding of parameter repeat and number. The function returns a duration corresponding to number times the execution of the main statement.

source on GitHub

mlprodict.testing.einsum.einsum_bench.einsum_benchmark(equation='abc,cd->abd', shape=30, perm=False, runtime='python', use_tqdm=False, number=5, repeat=5, opset=17)#

Investigates whether or not the decomposing einsum is faster.

Parameters:

  • equation – einsum equation to test

  • shape – an integer (all dimension gets the same size) or a list of shapes in a string separated with ;)

  • perm – check on permutation or all letter permutations

  • runtime – numpy, python, onnxruntime

  • use_tqdm – show progress

  • output – output file (usually a csv file or an excel file), it requires pandas

  • number – usual parameter to measure a function

  • repeat – usual parameter to measure a function

  • opset – target opset

Returns:

list of dictionaries as an iterator

source on GitHub