Skip to content

Parallel correlation calculation of big numpy arrays or pandas dataframes with NaNs and infs.

License

Notifications You must be signed in to change notification settings

bukson/nancorrmp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multiprocessing correlation calculation for Python

Build Status

nancorrmp is a small module for calculating correlations of big numpy arrays or pandas dataframes with NaNs and infs, using multiple cores. Default numpy.corrcoef method does not calculate correlations with input that contains NaNs and infs and pandas method pandas.DataFrame.corr is single thread only.

nancorrmp utilizes Pearson correlation calculation code from scipy, that is based on numpy instead of pandas cythonic backed. The multiprocessing is implemented by python multiprocessing module. nancorrmp uses pandas method of calculating correlations of arrays with NaNs and infs, that skips pair of observations when one of them is either Nan or +inf, or -inf. nancorrmp also can calculate result with p values, similar to scipy.pearsonr function.

Benchmarks are showing that with 4 cores, calculating correlation is faster with nancorrmp then with pandas even for 1200x1200 matrix. With 2 cores it is for 2400x2400. pandas single processed implementation is faster then using single process nancorrmp still for 5000x5000 matrix, so it is recommended to use nancorrmp with at least 2 cores.

Table of Content

Installation

pip install nancorrmp

Usage

import pandas as pd
import numpy as np
from nancorrmp.nancorrmp import NaNCorrMp
from pandas.testing import assert_frame_equal

np.random.seed(0)
random_dataframe = pd.DataFrame(np.random.rand(100, 100))
corr = NaNCorrMp.calculate(random_dataframe)
corr_pandas = random_dataframe.corr()
assert_frame_equal(corr, corr_pandas)
corr, p_value = NaNCorrMp.calculate_with_p_value(random_dataframe)

NaNCorrMp Methods

nancorrmp module has one static class named NaNCorrMp with 2 public methods and 1 type

ArrayLike = Union[pd.DataFrame, np.ndarray]

Type used to unify pd.DataFrame and np.ndarray.

NaNCorrMp.calculate(X: ArrayLike, n_jobs: int = -1, chunks: int = 500) -> ArrayLike

Calculates correlation matrix using Pearson correlation. n_jobs controls number of cores to use with default -1 which uses all available cores. chunks controls how many pairs of arrays are send to each process, 500 should be suitable for all purposes.

Returns output as the same type as input, if X is pd.Dataframe it will return pd.Dataframe, if X is np.ndarray it will return np.ndarray.

import pandas as pd
import numpy as np
from nancorrmp.nancorrmp import NaNCorrMp

np.random.seed(0)
random_dataframe = pd.DataFrame(np.random.rand(100, 100))
corr = NaNCorrMp.calculate(random_dataframe)

NaNCorrMp.calculate_with_p_value(X: ArrayLike, n_jobs: int = -1, chunks: int = 500) -> Tuple[ArrayLike, ArrayLike]

Calculates correlation matrix and p value matrix using Pearson correlation. n_jobs controls number of cores to use with default -1 which uses all available cores. chunks controls how many pairs of arrays are send to each process, 500 should be suitable for all purposes. Correlation and p value are the same as the result of using scipy.pearsonr, but it can be used with NaNs and infs and multiple cores.

Returns output as similar type as input, if X is pd.Dataframe it will return (pd.Dataframe, pd.Dataframe), if X is np.ndarray it will return (np.ndarray, np.ndarray).

import pandas as pd
import numpy as np
from nancorrmp.nancorrmp import NaNCorrMp

np.random.seed(0)
random_dataframe = pd.DataFrame(np.random.rand(100, 100))
corr, p_value = NaNCorrMp.calculate_with_p_value(random_dataframe)

Benchmark

Results can be reproduced by using test/test_benchmark_nancorrmp.py module

import pandas as pd
import numpy as np
from nancorrmp.nancorrmp import NaNCorrMp

np.random.seed(0)
random_dataframe = pd.DataFrame(np.random.rand(1200, 1200))

%timeit NaNCorrMp.calculate(random_dataframe, n_jobs=4, chunks=1000)
# 9.92 s ± 205 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit random_dataframe.corr()
# 10.4 s ± 56.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

random_dataframe = pd.DataFrame(np.random.rand(2400, 2400))

%timeit NaNCorrMp.calculate(random_dataframe, n_jobs=2, chunks=1000)
# 1min 26s ± 3.16 s per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit random_dataframe.corr()
# 1min 45s ± 3.58 s per loop (mean ± std. dev. of 7 runs, 1 loop each)

Test

test module contains test both for single core usage as for multiple cores. Tests asserts then the outuput of NaNCorrMp.calculate is the same as output of pandas.corr for the same data. Tests require scipy and can be run with the following command:

python setup.py test

Licencse

MIT License

Copyright (c) 2020 Michał Bukowski michal.bukowski@buksoft.pl

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.