Skip to main content

Do likelihood based parameter estimation using maximum likeihood and bayesian methods

Project description

docs/badges/tests-badge.svg docs/badges/coverage-badge.svg

dataprob was designed to allow scientists to easily fit user-defined models to experimental data. It allows maximum likelihood, bootstrap, and Bayesian analyses with a simple and consistent interface.

Design principles

  • ease of use: Users write a python function that describes their model, then load in their experimental data as a dataframe.

  • dataframe centric: Uses a pandas dataframe to specify parameter bounds, guesses, fixedness, and priors. Observed data can be passed in as a dataframe or numpy vector. All outputs are pandas dataframes.

  • consistent experience: Users can run maximum-likelihood, bootstrap resampling, or Bayesian MCMC analyses with an identical interface and nearly identical diagnostic outputs.

  • interpretable: Provides diagnostic plots and runs tests to validate fit results.

Simple example

The following code generates noisy linear data and uses dataprob to find the maximum likelihood estimate of its slope and intercept. Run on Google Colab.

import dataprob
import numpy as np

# Generate "experimental" linear data (slope = 5, intercept = 5.7) that has
# random noise on each point.
x_array = np.linspace(0,10,25)
noise = np.random.normal(loc=0,scale=0.5,size=x_array.shape)
y_obs = 5*x_array + 5.7 + noise

# 1. Define a linear model
def linear_model(m=1,b=1,x=[]):
    return m*x + b

# 2. Set up the analysis. 'method' can be "ml", "mcmc", or "bootstrap"
f = dataprob.setup(linear_model,
                   method="ml",
                   non_fit_kwargs={"x":x_array})

# 3. Fit the parameters of linear_model model to y_obs, assuming uncertainty
#    of 0.5 on each observed point.
f.fit(y_obs=y_obs,
      y_std=0.5)

# 4. Access results
fig = dataprob.plot_summary(f)
fig = dataprob.plot_corner(f)
print(f.fit_df)
print(f.fit_quality)

The plots will be:

data.plot_summary result data.plot_corner result

The f.fit_df dataframe will look something like:

index

name

estimate

std

low_95

high_95

prior_std

m

m

5.009

0.045

4.817

5.202

NaN

b

b

5.644

0.274

4.465

6.822

NaN

The f.fit_quality dataframe will look something like:

name

description

is_good

value

num_obs

number of observations

True

25.000

num_param

number of fit parameters

True

2.000

lnL

log likelihood

True

-18.761

chi2

chi^2 goodness-of-fit

True

0.241

reduced_chi2

reduced chi^2

True

1.192

mean0_resid

t-test for residual mean != 0

True

1.000

durbin-watson

Durbin-Watson test for correlated residuals

True

2.265

ljung-box

Ljung-Box test for correlated residuals

True

0.943

Installation

We recommend installing dataprob with pip:

pip install dataprob

To install from source and run tests:

git clone https://github.com/harmslab/dataprob.git
cd dataprob
pip install .

# to run test-suite
pytest --runslow

Examples

A good way to learn how to use the library is by working through examples. The following notebooks are included in the dataprob/examples/ directory. They are self-contained demonstrations in which dataprob is used to analyze various classes of experimental data. The links below launch each notebook in Google Colab:

Documentation

Full documentation is on readthedocs.

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

dataprob-0.9.4.tar.gz (50.1 kB view details)

Uploaded Source

Built Distribution

dataprob-0.9.4-py2.py3-none-any.whl (62.5 kB view details)

Uploaded Python 2 Python 3

File details

Details for the file dataprob-0.9.4.tar.gz.

File metadata

  • Download URL: dataprob-0.9.4.tar.gz
  • Upload date:
  • Size: 50.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/5.1.1 CPython/3.12.6

File hashes

Hashes for dataprob-0.9.4.tar.gz
Algorithm Hash digest
SHA256 ad29a3498e5e90f2cca7435e32beff665065eb41d938e534df601e4352c8b38b
MD5 9d95b7f92681c727951f9b320d631537
BLAKE2b-256 5e6168dcf223120b8d4feceff927e850f48aa54fc877ae81b8c3e3e656fb5378

See more details on using hashes here.

Provenance

File details

Details for the file dataprob-0.9.4-py2.py3-none-any.whl.

File metadata

  • Download URL: dataprob-0.9.4-py2.py3-none-any.whl
  • Upload date:
  • Size: 62.5 kB
  • Tags: Python 2, Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/5.1.1 CPython/3.12.6

File hashes

Hashes for dataprob-0.9.4-py2.py3-none-any.whl
Algorithm Hash digest
SHA256 58ad959ed297c9346ea27c3c5e84d41768b2e4a6c9f9b96ca769e73f5cd98223
MD5 d123c5bd79e65a379bb67469d8bc83d9
BLAKE2b-256 b6e060a95ddf57e74c18fb56242cb651826f281d849d04ffde090519c4fbed11

See more details on using hashes here.

Provenance

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page