Prediction LatencyΒΆ
This is an example showing the prediction latency of various scikit-learn estimators.
The goal is to measure the latency one can expect when doing predictions either in bulk or atomic (i.e. one by one) mode.
The plots represent the distribution of the prediction latency as a boxplot.
Script output:
Benchmarking SGDRegressor(alpha=0.01, epsilon=0.1, eta0=0.01, fit_intercept=True,
l1_ratio=0.25, learning_rate='invscaling', loss='squared_loss',
n_iter=5, penalty='elasticnet', power_t=0.25, random_state=None,
shuffle=False, verbose=0, warm_start=False)
Benchmarking RandomForestRegressor(bootstrap=True, compute_importances=None,
criterion='mse', max_depth=None, max_features='auto',
max_leaf_nodes=None, min_density=None, min_samples_leaf=1,
min_samples_split=2, n_estimators=10, n_jobs=1, oob_score=False,
random_state=None, verbose=0)
Benchmarking SVR(C=1.0, cache_size=200, coef0=0.0, degree=3, epsilon=0.1, gamma=0.0,
kernel='rbf', max_iter=-1, probability=False, random_state=None,
shrinking=True, tol=0.001, verbose=False)
agg_filter: unknown
alpha: float (0.0 transparent through 1.0 opaque)
animated: [True | False]
axes: an :class:`~matplotlib.axes.Axes` instance
backgroundcolor: any matplotlib color
bbox: rectangle prop dict
clip_box: a :class:`matplotlib.transforms.Bbox` instance
clip_on: [True | False]
clip_path: [ (:class:`~matplotlib.path.Path`, :class:`~matplotlib.transforms.Transform`) | :class:`~matplotlib.patches.Patch` | None ]
color: any matplotlib color
contains: a callable function
family or fontfamily or fontname or name: [ FONTNAME | 'serif' | 'sans-serif' | 'cursive' | 'fantasy' | 'monospace' ]
figure: a :class:`matplotlib.figure.Figure` instance
fontproperties or font_properties: a :class:`matplotlib.font_manager.FontProperties` instance
gid: an id string
horizontalalignment or ha: [ 'center' | 'right' | 'left' ]
label: any string
linespacing: float (multiple of font size)
lod: [True | False]
multialignment: ['left' | 'right' | 'center' ]
path_effects: unknown
picker: [None|float|boolean|callable]
position: (x,y)
rasterized: [True | False | None]
rotation: [ angle in degrees | 'vertical' | 'horizontal' ]
rotation_mode: unknown
size or fontsize: [ size in points | 'xx-small' | 'x-small' | 'small' | 'medium' | 'large' | 'x-large' | 'xx-large' ]
snap: unknown
stretch or fontstretch: [ a numeric value in range 0-1000 | 'ultra-condensed' | 'extra-condensed' | 'condensed' | 'semi-condensed' | 'normal' | 'semi-expanded' | 'expanded' | 'extra-expanded' | 'ultra-expanded' ]
style or fontstyle: [ 'normal' | 'italic' | 'oblique']
text: string or anything printable with '%s' conversion.
transform: :class:`~matplotlib.transforms.Transform` instance
url: a url string
variant or fontvariant: [ 'normal' | 'small-caps' ]
verticalalignment or va or ma: [ 'center' | 'top' | 'bottom' | 'baseline' ]
visible: [True | False]
weight or fontweight: [ a numeric value in range 0-1000 | 'ultralight' | 'light' | 'normal' | 'regular' | 'book' | 'medium' | 'roman' | 'semibold' | 'demibold' | 'demi' | 'bold' | 'heavy' | 'extra bold' | 'black' ]
x: float
y: float
zorder: any number
agg_filter: unknown
alpha: float (0.0 transparent through 1.0 opaque)
animated: [True | False]
axes: an :class:`~matplotlib.axes.Axes` instance
backgroundcolor: any matplotlib color
bbox: rectangle prop dict
clip_box: a :class:`matplotlib.transforms.Bbox` instance
clip_on: [True | False]
clip_path: [ (:class:`~matplotlib.path.Path`, :class:`~matplotlib.transforms.Transform`) | :class:`~matplotlib.patches.Patch` | None ]
color: any matplotlib color
contains: a callable function
family or fontfamily or fontname or name: [ FONTNAME | 'serif' | 'sans-serif' | 'cursive' | 'fantasy' | 'monospace' ]
figure: a :class:`matplotlib.figure.Figure` instance
fontproperties or font_properties: a :class:`matplotlib.font_manager.FontProperties` instance
gid: an id string
horizontalalignment or ha: [ 'center' | 'right' | 'left' ]
label: any string
linespacing: float (multiple of font size)
lod: [True | False]
multialignment: ['left' | 'right' | 'center' ]
path_effects: unknown
picker: [None|float|boolean|callable]
position: (x,y)
rasterized: [True | False | None]
rotation: [ angle in degrees | 'vertical' | 'horizontal' ]
rotation_mode: unknown
size or fontsize: [ size in points | 'xx-small' | 'x-small' | 'small' | 'medium' | 'large' | 'x-large' | 'xx-large' ]
snap: unknown
stretch or fontstretch: [ a numeric value in range 0-1000 | 'ultra-condensed' | 'extra-condensed' | 'condensed' | 'semi-condensed' | 'normal' | 'semi-expanded' | 'expanded' | 'extra-expanded' | 'ultra-expanded' ]
style or fontstyle: [ 'normal' | 'italic' | 'oblique']
text: string or anything printable with '%s' conversion.
transform: :class:`~matplotlib.transforms.Transform` instance
url: a url string
variant or fontvariant: [ 'normal' | 'small-caps' ]
verticalalignment or va or ma: [ 'center' | 'top' | 'bottom' | 'baseline' ]
visible: [True | False]
weight or fontweight: [ a numeric value in range 0-1000 | 'ultralight' | 'light' | 'normal' | 'regular' | 'book' | 'medium' | 'roman' | 'semibold' | 'demibold' | 'demi' | 'bold' | 'heavy' | 'extra bold' | 'black' ]
x: float
y: float
zorder: any number
benchmarking with 100 features
benchmarking with 250 features
benchmarking with 500 features
example run in 4.15s
Python source code: plot_prediction_latency.py
# Authors: Eustache Diemert <eustache@diemert.fr>
# License: BSD 3 clause
from __future__ import print_function
from collections import defaultdict
import time
import gc
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import scoreatpercentile
from sklearn.datasets.samples_generator import make_regression
from sklearn.ensemble.forest import RandomForestRegressor
from sklearn.linear_model.ridge import Ridge
from sklearn.linear_model.stochastic_gradient import SGDRegressor
from sklearn.svm.classes import SVR
def _not_in_sphinx():
# Hack to detect whether we are running by the sphinx builder
return '__file__' in globals()
def atomic_benchmark_estimator(estimator, X_test, verbose=False):
"""Measure runtime prediction of each instance."""
n_instances = X_test.shape[0]
runtimes = np.zeros(n_instances, dtype=np.float)
for i in range(n_instances):
instance = X_test[i, :]
start = time.time()
estimator.predict(instance)
runtimes[i] = time.time() - start
if verbose:
print("atomic_benchmark runtimes:", min(runtimes), scoreatpercentile(
runtimes, 50), max(runtimes))
return runtimes
def bulk_benchmark_estimator(estimator, X_test, n_bulk_repeats, verbose):
"""Measure runtime prediction of the whole input."""
n_instances = X_test.shape[0]
runtimes = np.zeros(n_bulk_repeats, dtype=np.float)
for i in range(n_bulk_repeats):
start = time.time()
estimator.predict(X_test)
runtimes[i] = time.time() - start
runtimes = np.array(list(map(lambda x: x / float(n_instances), runtimes)))
if verbose:
print("bulk_benchmark runtimes:", min(runtimes), scoreatpercentile(
runtimes, 50), max(runtimes))
return runtimes
def benchmark_estimator(estimator, X_test, n_bulk_repeats=30, verbose=False):
"""
Measure runtimes of prediction in both atomic and bulk mode.
Parameters
----------
estimator : already trained estimator supporting `predict()`
X_test : test input
n_bulk_repeats : how many times to repeat when evaluating bulk mode
Returns
-------
atomic_runtimes, bulk_runtimes : a pair of `np.array` which contain the
runtimes in seconds.
"""
atomic_runtimes = atomic_benchmark_estimator(estimator, X_test, verbose)
bulk_runtimes = bulk_benchmark_estimator(estimator, X_test, n_bulk_repeats,
verbose)
return atomic_runtimes, bulk_runtimes
def generate_dataset(n_train, n_test, n_features, noise=0.1, verbose=False):
"""Generate a regression dataset with the given parameters."""
if verbose:
print("generating dataset...")
X, y, coef = make_regression(n_samples=n_train + n_test,
n_features=n_features, noise=noise, coef=True)
X_train = X[:n_train]
y_train = y[:n_train]
X_test = X[n_train:]
y_test = y[n_train:]
idx = np.arange(n_train)
np.random.seed(13)
np.random.shuffle(idx)
X_train = X_train[idx]
y_train = y_train[idx]
std = X_train.std(axis=0)
mean = X_train.mean(axis=0)
X_train = (X_train - mean) / std
X_test = (X_test - mean) / std
std = y_train.std(axis=0)
mean = y_train.mean(axis=0)
y_train = (y_train - mean) / std
y_test = (y_test - mean) / std
gc.collect()
if verbose:
print("ok")
return X_train, y_train, X_test, y_test
def boxplot_runtimes(runtimes, pred_type, configuration):
"""
Plot a new `Figure` with boxplots of prediction runtimes.
Parameters
----------
runtimes : list of `np.array` of latencies in micro-seconds
cls_names : list of estimator class names that generated the runtimes
pred_type : 'bulk' or 'atomic'
"""
fig, ax1 = plt.subplots(figsize=(10, 6))
bp = plt.boxplot(runtimes, )
cls_infos = ['%s\n(%d %s)' % (estimator_conf['name'],
estimator_conf['complexity_computer'](
estimator_conf['instance']),
estimator_conf['complexity_label']) for
estimator_conf in configuration['estimators']]
xtick_names = plt.setp(ax1, xticklabels=cls_infos)
plt.setp(xtick_names)
plt.setp(bp['boxes'], color='black')
plt.setp(bp['whiskers'], color='black')
plt.setp(bp['fliers'], color='red', marker='+')
ax1.yaxis.grid(True, linestyle='-', which='major', color='lightgrey',
alpha=0.5)
ax1.set_axisbelow(True)
ax1.set_title('Prediction Time per Instance - %s, %d feats.' % (
pred_type.capitalize(),
configuration['n_features']))
ax1.set_ylabel('Prediction Time (us)')
plt.show()
def benchmark(configuration):
"""Run the whole benchmark."""
X_train, y_train, X_test, y_test = generate_dataset(
configuration['n_train'], configuration['n_test'],
configuration['n_features'])
stats = {}
for estimator_conf in configuration['estimators']:
print("Benchmarking", estimator_conf['instance'])
estimator_conf['instance'].fit(X_train, y_train)
gc.collect()
a, b = benchmark_estimator(estimator_conf['instance'], X_test)
stats[estimator_conf['name']] = {'atomic': a, 'bulk': b}
cls_names = [estimator_conf['name'] for estimator_conf in configuration[
'estimators']]
runtimes = [1e6 * stats[clf_name]['atomic'] for clf_name in cls_names]
boxplot_runtimes(runtimes, 'atomic', configuration)
runtimes = [1e6 * stats[clf_name]['bulk'] for clf_name in cls_names]
boxplot_runtimes(runtimes, 'bulk (%d)' % configuration['n_test'],
configuration)
def n_feature_influence(estimators, n_train, n_test, n_features, percentile):
"""
Estimate influence of the number of features on prediction time.
Parameters
----------
estimators : dict of (name (str), estimator) to benchmark
n_train : nber of training instances (int)
n_test : nber of testing instances (int)
n_features : list of feature-space dimensionality to test (int)
percentile : percentile at which to measure the speed (int [0-100])
Returns:
--------
percentiles : dict(estimator_name,
dict(n_features, percentile_perf_in_us))
"""
percentiles = defaultdict(defaultdict)
for n in n_features:
print("benchmarking with %d features" % n)
X_train, y_train, X_test, y_test = generate_dataset(n_train, n_test, n)
for cls_name, estimator in estimators.items():
estimator.fit(X_train, y_train)
gc.collect()
runtimes = bulk_benchmark_estimator(estimator, X_test, 30, False)
percentiles[cls_name][n] = 1e6 * scoreatpercentile(runtimes,
percentile)
return percentiles
def plot_n_features_influence(percentiles, percentile):
fig, ax1 = plt.subplots(figsize=(10, 6))
colors = ['r', 'g', 'b']
for i, cls_name in enumerate(percentiles.keys()):
x = np.array(sorted([n for n in percentiles[cls_name].keys()]))
y = np.array([percentiles[cls_name][n] for n in x])
plt.plot(x, y, color=colors[i], )
ax1.yaxis.grid(True, linestyle='-', which='major', color='lightgrey',
alpha=0.5)
ax1.set_axisbelow(True)
ax1.set_title('Evolution of Prediction Time with #Features')
ax1.set_xlabel('#Features')
ax1.set_ylabel('Prediction Time at %d%%-ile (us)' % percentile)
plt.show()
def benchmark_throughputs(configuration, duration_secs=0.1):
"""benchmark throughput for different estimators."""
X_train, y_train, X_test, y_test = generate_dataset(
configuration['n_train'], configuration['n_test'],
configuration['n_features'])
throughputs = dict()
for estimator_config in configuration['estimators']:
estimator_config['instance'].fit(X_train, y_train)
start_time = time.time()
n_predictions = 0
while (time.time() - start_time) < duration_secs:
estimator_config['instance'].predict(X_test[0])
n_predictions += 1
throughputs[estimator_config['name']] = n_predictions / duration_secs
return throughputs
def plot_benchmark_throughput(throughputs, configuration):
fig, ax = plt.subplots(figsize=(10, 6))
colors = ['r', 'g', 'b']
cls_infos = ['%s\n(%d %s)' % (estimator_conf['name'],
estimator_conf['complexity_computer'](
estimator_conf['instance']),
estimator_conf['complexity_label']) for
estimator_conf in configuration['estimators']]
cls_values = [throughputs[estimator_conf['name']] for estimator_conf in
configuration['estimators']]
plt.bar(range(len(throughputs)), cls_values, width=0.5, color=colors)
ax.set_xticks(np.linspace(0.25, len(throughputs) - 0.75, len(throughputs)))
ax.set_xticklabels(cls_infos, fontsize=10)
ymax = max(cls_values) * 1.2
ax.set_ylim((0, ymax))
ax.set_ylabel('Throughput (predictions/sec)')
ax.set_title('Prediction Throughput for different estimators (%d '
'features)' % configuration['n_features'])
plt.show()
###############################################################################
# main code
start_time = time.time()
# benchmark bulk/atomic prediction speed for various regressors
configuration = {
'n_train': int(1e3),
'n_test': int(1e2),
'n_features': int(1e2),
'estimators': [
{'name': 'Linear Model',
'instance': SGDRegressor(penalty='elasticnet', alpha=0.01,
l1_ratio=0.25, fit_intercept=True),
'complexity_label': 'non-zero coefficients',
'complexity_computer': lambda clf: np.count_nonzero(clf.coef_)},
{'name': 'RandomForest',
'instance': RandomForestRegressor(),
'complexity_label': 'estimators',
'complexity_computer': lambda clf: clf.n_estimators},
{'name': 'SVR',
'instance': SVR(kernel='rbf'),
'complexity_label': 'support vectors',
'complexity_computer': lambda clf: len(clf.support_vectors_)},
]
}
benchmark(configuration)
# benchmark n_features influence on prediction speed
percentile = 90
percentiles = n_feature_influence({'ridge': Ridge()},
configuration['n_train'],
configuration['n_test'],
[100, 250, 500], percentile)
plot_n_features_influence(percentiles, percentile)
# benchmark throughput
throughputs = benchmark_throughputs(configuration)
plot_benchmark_throughput(throughputs, configuration)
stop_time = time.time()
print("example run in %.2fs" % (stop_time - start_time))
Total running time of the example: 4.23 seconds ( 0 minutes 4.23 seconds)