csv-stats
Python package for rapid hypothesis testing on CSV files with data in long table format, which are common in the life sciences. Test results are saved to PDF as a rendered JSON string, or can be returned as a Python dictionary.
Installation
pip install "csv-stats[full]" # hypothesis tests + PDF reports
pip install csv-stats # descriptive statistics only (numpy, pandas, scipy)
Descriptive statistics
csvstats.descriptive is the one definition of every descriptive formula: NaN
skipped (counted in n_missing), SD with n - 1, linear quartiles, and a
t-based 95% CI of the mean. Fewer than two values give NaN, not 0.
from csvstats.descriptive import summarize, ci_interval, ci_half_width
summarize([4.1, 5.0, 6.2, None]) # n=3, n_missing=1, mean, sd, sem, ci_low, ci_high, median, q1, q3, iqr, skewness, kurtosis, ...
ci_interval([4.1, 5.0, 6.2]) # (low, high); also a seaborn errorbar= callable
ci_half_width(sd=1.2, n=5) # t(0.975, 4) * 1.2 / sqrt(5)
Linear mixed models (0.4.1)
csvstats.lmm uses every observation: a subject missing some within-subject
levels is kept, not dropped (a repeated-measures ANOVA or paired t-test would
have to drop it). REML, sum-to-zero coding. The random effects default to an
intercept per subject. random= takes any lme4-style structure:
- random slopes, correlated or uncorrelated (
||) - crossed groupings, e.g. subject + item
- nested and interaction groupings, e.g.
site:subject
Numeric covariates= are crossed with the factors. Term tests are type III,
as in lmerTest and SAS. So a factor crossed with a covariate is tested at
covariate = 0, and a type3_at_covariate_zero warning says so. Centre the
covariate to test the factor at the covariate's mean.
csvstats.lmm_formula reads and writes the lme4 spelling of these models
(parse("y ~ group * time * x + (1 + x | subject)"), render(...)), with
errors that point at the offending character. It accepts what lmm can fit:
the fixed part is ONE crossing (a * b * x, or 1), and random terms are
(… | g), (… || g), a:b and a/b. A model with no more observations
than random effects is refused before fitting, as lme4 does.
lmm_pairwise takes the same covariates= and compares levels with each
covariate at its mean (emmeans' reference grid), or at the value you give
in covariates_at={"age": 65}. The record's details["covariates_at"]
states the value used. So with a covariate, lmm's F for a factor (at 0)
and lmm_pairwise's contrasts (at the mean) answer different questions,
exactly as lmerTest's anova and emmeans do in R.
The fit is checked against lme4 on 225 sampled scenarios over 9
random-effect structures (tests/reference/scenarios). That covers the
criterion itself at fixed θ, the estimates, θ and the Satterthwaite df.
Tests use Satterthwaite degrees of freedom, as R's lmerTest does: t(df) per
coefficient, a type-III F(q, ddf) per term, and t(df) per pairwise contrast. The
pairwise contrasts come from the full model (emmeans-style). Everything is checked
against lme4 / lmerTest / emmeans on 75 reference datasets (tests/reference).
If a df cannot be computed, the Wald test is reported, labelled df_method: "wald".
from csvstats.lmm import lmm, lmm_pairwise
from csvstats.core import assumption_checks
lmm(df, "step_length", within=["session", "speed"], subject="subject") # one F(q, ddf) per term
lmm_pairwise(df, "step_length", within=["session", "speed"], subject="subject", factor="session",
levels={"session": ["post", "pre"]}) # post − pre within each speed, t(df)
assumption_checks(df, "step_length", within=["session"], subject="subject") # Shapiro–Wilk on residuals (+ Levene)
# Random slopes and crossed effects: y ~ cond * x + (1 + x | subject) + (1 | item)
lmm(df, "rt", within=["cond"], subject="subject", covariates=["x"],
random=[{"group": "subject", "terms": ["(Intercept)", "x"]},
{"group": "item"}])
Examples
All code examples below use the following constants:
DATA_PATH = "path/to/data.csv" # Path to your CSV file
DATA_COLUMN = 'values' # The column to run the hypothesis tests on
GROUP_COLUMN = 'groups' # Grouping variable (i.e. statistical factor)
REPEATED_MEASURES_COLUMN = 'subject_id' # Column indicating repeated measures (e.g. subject IDs)
ANOVA
One way ANOVA is currently supported. They include tests of homogeneity of variance and normality of residuals. Repeated measures ANOVA is also supported, including tests of sphericity.
NOTE: Two- and three-way ANOVA support is planned, but not yet implemented.
from csv_stats.anova import anova1way
# One way ANOVA, independent samples
result_anova1way = anova1way(DATA_PATH,
GROUP_COLUMN,
DATA_COLUMN,
filename = "anova1way_results.pdf", # Default save name. Enter `None` to not save.
render_plot = False # For speed, by default no plots are generated
)
# One way ANOVA, repeated measures
result_anova1way_rm = anova1way(DATA_PATH,
GROUP_COLUMN,
DATA_COLUMN,
repeated_measures_column = REPEATED_MEASURES_COLUMN,
filename = "anova1way_results.pdf", # Default save name. Enter `None` to not save.
render_plot = False # For speed, by default no plots are generated
)
t-test
Both independent samples (one and two samples) and paired samples t-tests are supported. They include tests of homogeneity of variance and normality of residuals.
from csv_stats.ttest import ttest_ind, ttest_dep
# Independent samples t-test
# Two sample when the GROUP_COLUMN has two groups
# One smaple when the GROUP_COLUMN has one group
result_ttest_ind = ttest_ind(DATA_PATH,
GROUP_COLUMN,
DATA_COLUMN,
popmean = 0, # Test against a population mean of 0 (default)
filename = "ttest_ind_results.pdf", # Default save name. Enter `None` to not save.
render_plot = False # For speed, by default no plots are generated
)
# Paired samples t-test
result_ttest_rel = ttest_dep(DATA_PATH,
GROUP_COLUMN,
DATA_COLUMN,
repeated_measures_column = REPEATED_MEASURES_COLUMN,
filename = "ttest_dep_results.pdf", # Default save name. Enter `None` to not save.
render_plot = False # For speed, by default no plots are generated
)
Multiple Data Columns
If you have multiple columns of data that you want to run the same test on, you can specify the data_column argument as "_". This will automatically loop over all columns. Note that to save these results to a file, the filename should be an f-string containing f"{data_column}". The test function will replace data_column with the column name in the file name.
Metadata
Release files for csv-stats 0.4.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| csv_stats-0.4.2.tar.gz | 1.1 MB | Without provenance | Details for csv_stats-0.4.2.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| csv_stats-0.4.2-py3-none-any.whl | Python 3 | none | any | Details for csv_stats-0.4.2-py3-none-any.whl |
Total release size: 1.2 MB
Release files / csv_stats-0.4.2.tar.gz
| Download URL | csv_stats-0.4.2.tar.gz |
|---|---|
| Size | 1.1 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
903f53a86744ecb815b4045c66c5d5af69cf0ee63e8921f82955bc933e951e31
|
|
BLAKE2b-256 checksum How to use checksums |
70bdccfbce97ca053cf806f088c09f856070543346fd322baa4d7d5309097ca8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.11
|
Release files / csv_stats-0.4.2-py3-none-any.whl
| Download URL | csv_stats-0.4.2-py3-none-any.whl |
|---|---|
| Size | 60.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
68a08dc44102044b9b7c34505c32c969e31053742a30b2da37819bb07e985e5a
|
|
BLAKE2b-256 checksum How to use checksums |
2740b559e6f8f08be1c6a052abd59a42cd6869f8e3d5ecc998a60d4eaa8d57fd
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.11.11
|
Security for csv-stats 0.4.2
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.