A fast implementation of bootstrapping supporting multi-columns data.
Project description
Strapping
Strapping is a library containing a fast implementation of bootstrapping sampling algorithm. Along the sampling algorithms you will find a set of helper functions used to compute basic statistics useful in bootstrapping-based analysis.
Library supports:
- single variable sampling
- multi-column variable sampling
- A/B test difference sampling
Installing
Strapping can be installed via pip from PyPI.
pip install strapping
Testing
Tu run tests for the package use tox
:
tox
Example
Sample single variable
In this example we will use a bootstrapping algorithm to sample a distribution of mean and std. deviation of the given dataset.
Sample means using bootstrapping
Import bootstrap
and stats
module.
bootstrap
contains bootstrapping algorithms,stats
contains helpers for computing basic statistics (e.g. confidence intervals).
from strapping import bootstrap, stats
Generate sample data using normal distribution:
X = np.random.normal(0, 1, size=100).reshape(-1, 1)
Sample a vector containing possible means for given dataset:
mu_sampled = bootstrap.sample(X, iterations=1000, aggrfunc=np.mean)
std_sampled = bootstrap.sample(X, iterations=1000, aggrfunc=np.std)
We can check output values:
>>> np.mean(mu_sampled), np.mean(std_sampled)
(-0.028259915654785906, 1.0099170040429664)
Compute confidence intervals
Now we will compute confidence intervals based on sampled values. This works for both single values and multi-column variables. By default, confidence interval will three values: (5th quantile, mean, 95th quantile).
q05, mean, q95 = stats.confidence_intervals(mu_sampled)
We can check output values:
>>> q05
array([-0.15844911])
>>> mean
array([-0.01509199])
>>> q95
array([0.12659994])
Sample multi-column variables
In this example we will test using bootstrapping for data containing multiple columns.
Generate data containing multiple columns:
X = np.array([
np.random.normal(0, 1, size=100),
np.random.normal(10, 5, size=100),
np.random.normal(-20, 5, size=100),
]).T
Import bootstrap
module:
from strapping import bootstrap
Sample mean for given dataset:
mu_sampled = bootstrap.sample(X, iterations=1000, aggrfunc=np.mean)
We can check output values:
>>> mu_sampled.mean(axis=0)
array([ -0.06588892, 9.97571153, -19.187514 ])
A/B test difference between two variables
In this example we will test using bootstrapping to sample a difference between two given datasets. Then, we will use sampled values to compute percentage confidence intervals for the difference.
Sample means using bootstrapping
Generate data containing multiple columns:
X1 = np.random.normal(5, 2, size=100).reshape(-1, 1)
X2 = np.random.normal(6, 2, size=100).reshape(-1, 1)
Import bootstrap
and stats
modules:
from strapping import bootstrap, stats
Sample mean for given dataset:
mu_sampled = bootstrap.sample_diffs(X1, X2, iterations=1000, aggrfunc=np.mean)
We can check output values:
>>> mu_sampled.mean()
-1.2875678613575356
Compute confidence intervals
Now we will compute both confidence intervals and percentage confidence intervals based on sampled values.
>>> stats.confidence_intervals(mu_sampled)
(array([-1.77019123]), array([-1.28756786]), array([-0.79820009]))
Percentage confidence intervals are computed as a percentage difference between sampled values and the mean value of a provided reference (control dataset).
>>> stats.percentage_confidence_intervals(mu_sampled, X1.mean())
(array([-0.36300107]), array([-0.26403278]), array([-0.16368146]))
Other
Compute Cohen's d
Using strapping
you can easily compute bootstrapped value of Cohen's d,
which is often used for a metric of measuring the effect size.
To do so first compute the difference between two datasets:
diff_sampled = bootstrap.sample_diffs(X1, X2, iterations=1000, aggrfunc=np.mean)
Then, compute the pooled standard deviation using a helper function and finally compute Cohen's d value:
from strapping.stats import pooled_std
pstd = pooled_std(X1, X2)
cohensd = diff_sampled / pstd
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file strapping-0.1.5.tar.gz
.
File metadata
- Download URL: strapping-0.1.5.tar.gz
- Upload date:
- Size: 18.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.9.5
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 87777f4cc6d5a64f784578aa350db96ea6ab078da1a6e17e1c0275c0d0d86978 |
|
MD5 | 76356a62bca7f334059db0257948b9a0 |
|
BLAKE2b-256 | b221169791292b1f0ac9195fc6c5c12e8141f1c5ff791afc8e2040431181baa0 |
File details
Details for the file strapping-0.1.5-py3-none-any.whl
.
File metadata
- Download URL: strapping-0.1.5-py3-none-any.whl
- Upload date:
- Size: 17.7 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.9.5
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 563f23070d69fcd2df44fef143b24eacc09113f1a577459520886651c29cc21a |
|
MD5 | b6d90a1c2fbfab8adb07c4d6f1bdad0f |
|
BLAKE2b-256 | 57e564e93e2fe7827e531d46f8dcac009748cbc85667672f60834571f92efe0a |