Useful functions

Links: notebook, html, PDF, python, slides, GitHub

Benchmarks measures performances. It can be useful to know in which context. These are some functions to know better about it.

from jyquickhelper import add_notebook_menu
add_notebook_menu()

%matplotlib inline

Machine

from pymlbenchmark.context import machine_information
import numpy
machine_information(['numpy', 'pandas', 'onnx'])

[{'name': 'date', 'version': '2019-03-07'},
 {'name': 'python',
  'value': '3.7.2 (tags/v3.7.2:9a3ffc0492, Dec 23 2018, 23:09:28) [MSC v.1916 64 bit (AMD64)]'},
 {'name': 'platform', 'value': 'win32'},
 {'name': 'OS', 'value': 'Windows-10-10.0.17134-SP0'},
 {'name': 'machine', 'value': 'AMD64'},
 {'name': 'processor',
  'value': 'Intel64 Family 6 Model 78 Stepping 3, GenuineIntel'},
 {'name': 'release', 'value': '10'},
 {'name': 'architecture', 'value': ('64bit', 'WindowsPE')},
 {'name': 'numpy', 'version': '1.16.2'},
 {'name': 'onnx', 'version': 'not-imported'},
 {'name': 'pandas', 'version': '0.24.1'}]

import pandas
pandas.DataFrame(machine_information(['numpy', 'pandas', 'onnx']))

	name	value	version
0	date	NaN	2019-03-07
1	python	3.7.2 (tags/v3.7.2:9a3ffc0492, Dec 23 2018, 23...	NaN
2	platform	win32	NaN
3	OS	Windows-10-10.0.17134-SP0	NaN
4	machine	AMD64	NaN
5	processor	Intel64 Family 6 Model 78 Stepping 3, GenuineI...	NaN
6	release	10	NaN
7	architecture	(64bit, WindowsPE)	NaN
8	numpy	NaN	1.16.2
9	onnx	NaN	not-imported
10	pandas	NaN	0.24.1

One benchmark

from pymlbenchmark.datasets import experiment_results
df = experiment_results('onnxruntime_LogisticRegression')
df.head()

	N	count	dim	fit_intercept	lib	lower	max	max3	mean	median	method	min	min3	repeat	upper
0	1	100	1	True	skl	-0.000014	0.000392	0.000128	0.000071	0.000052	predict	0.000038	0.000039	100	0.000155
1	1	100	1	True	ort	-0.000008	0.000185	0.000042	0.000025	0.000023	predict	0.000022	0.000022	100	0.000058
2	1	100	1	True	skl	0.000028	0.000121	0.000083	0.000051	0.000047	predict_proba	0.000046	0.000046	100	0.000074
3	1	100	1	True	ort	0.000014	0.000062	0.000045	0.000026	0.000024	predict_proba	0.000023	0.000024	100	0.000037
4	10	100	1	True	skl	0.000020	0.000111	0.000077	0.000043	0.000039	predict	0.000038	0.000039	100	0.000066

from pymlbenchmark.plotting import plot_bench_results
plot_bench_results(df, row_cols='N', col_cols='method',
                   x_value='dim', hue_cols='fit_intercept',
                   title="LogisticRegression\nBenchmark scikit-learn / onnxruntime");