PyBridge - MATLAB Python Scientific Computing Toolbox

PyBridge is a MATLAB toolbox that provides a clean interface to Python scientific computing libraries (scipy, statsmodels, linearmodels, econml), enabling users to call Python functions via the py. syntax while automatically handling data conversion and result parsing.

Features

Environment Management: Automatic Python environment detection and configuration
Data Conversion: Bidirectional MATLAB-Python data conversion
Result Parsing: Convert Python return objects to MATLAB-friendly formats
Error Handling: Unified exception handling with MATLAB-style error messages
Complete Wrappers:
- StatsmodelsWrapper - OLS, WLS, GLM, Logistic, Probit, MNLogit, Ordered, NegBin, ZINB, VAR, ARIMA, and more
- LinearmodelsWrapper - PanelOLS, RandomEffects, BetweenOLS, PooledOLS, FirstDifference, IV2SLS/LIML/GMM, Hausman test
- EconmlWrapper - DML, DR Learner, S/T/X Learner, Causal Forest, sensitivity analysis
- ScipyWrapper - stats, optimize, signal, linalg, integrate
Supporting Components:
- CovarianceTypes - HAC, HC0-HC3, Clustered, Multi-way Clustered, MLE Robust
- ResultParser - Automatic parsing of Python results to MATLAB structures
- DataConverter - Bidirectional MATLAB-Python data conversion with py.None handling
- ErrorHandler - Unified exception handling with MATLAB-style error messages
- PyBridgeConfig - Environment configuration and package verification

Advanced Econometrics Support

Discrete Choice Models: Multinomial Logit, Ordered Logit/Probit, Poisson, Negative Binomial, Zero-Inflated Negative Binomial (ZINB)
Robust Standard Errors: HC0-HC3 heteroskedasticity-robust standard errors
HAC Standard Errors: Newey-West standard errors (handling autocorrelation and heteroskedasticity)
Clustered Standard Errors: Single-way clustering, multi-way clustering (e.g., firm x year two-way clustering)
Panel Data: Fixed effects, random effects, two-way fixed effects with clustered standard errors
Time Series: VAR models, ARIMA, Durbin-Watson test, unit root tests
Generalized Models: GLM (Gaussian, Binomial, Poisson, Gamma, InverseGaussian families)

Test Status

137 tests, 135 passing, 0 failures, 2 skipped (6 test suites)

Test Suite	Description	Status
AcademicCorrectionsTest	Academic notation and statistical corrections	✓ Pass
MLogitHACSpecializedTest	Multinomial Logit with HAC standard errors	✓ Pass
PyBridgeConsistencyTest	Core functionality and API consistency	✓ Pass
PanelCausalTest	Panel data analysis and causal inference (EconmlWrapper + LinearmodelsWrapper)	✓ Pass
StatsParserCovTest	Statistical functions, result parsing and covariance types (ScipyStats + ResultParser + CovarianceTypes)	✓ Pass
StatsmodelsExtendedTest	Extended econometric models (WLS, GLM, NegBin, ZINB, VAR, Durbin-Watson, etc.)	✓ Pass

Quick Start

1. Requirements

MATLAB R2020b or higher
Python 3.7+ (miniconda/anaconda recommended)
- Python 3.13 Compatibility: Fully tested with MATLAB R2025b, all features working correctly (though not officially supported by MathWorks yet)
- linearmodels v7.0 Compatibility: Updated for new PanelOLS API constructor syntax

Required Python packages:

pip install scipy statsmodels linearmodels econml numpy pandas scikit-learn

2. Configure Environment

% Add toolbox to path (use the actual path where you cloned this repo)
addpath(genpath('path/to/MATLAB-Python-Econometrics-Bridge'));  % Or use startup.m

% Initialize environment
config = pyBridge.PyBridgeConfig();
config.initialize();
config.verifyAll();  % Verify all packages are installed

3. Basic Usage

scipy Statistics

% Create statistics instance
stats = pyBridge.internal.ScipyStats();

% Normal distribution
x = linspace(-3, 3, 100);
pdf = stats.normPDF(x, 0, 1);
cdf = stats.normCDF(x, 0, 1);

% t-test
data1 = randn(100, 1);
data2 = randn(100, 1) + 0.5;
result = stats.tTest(data1, data2);
fprintf('t-statistic: %.4f, p-value: %.4f\n', result.tStatistic, result.pValue);

% Normality test
normality = stats.normalityTest(data1, 'shapiro');
fprintf('Is normal: %s\n', ternary(normality.isNormal, 'Yes', 'No'));

scipy Optimization

% Create optimization instance
opt = pyBridge.internal.ScipyOptimize();

% Minimize function
fun = @(x) x(1)^2 + x(2)^2;
result = opt.minimize(fun, [1; 1]);
fprintf('Minimum: %.6f, Location: [%.6f, %.6f]\n', ...
    result.fun, result.x(1), result.x(2));

% Curve fitting
xData = linspace(0, 10, 50);
yData = 2.5 * xData + 1.5 + randn(size(xData)) * 0.5;

fitFun = @(x, a, b) a * x + b;
fitResult = opt.curveFit(fitFun, xData, yData);
fprintf('Fitted parameters: a=%.4f, b=%.4f\n', fitResult.parameters);

scipy Signal Processing

% Create signal processing instance
sig = pyBridge.internal.ScipySignal();

% Design Butterworth filter
[b, a] = sig.butter(4, 0.1, 'low');

% Filter signal
fs = 1000; % Sampling rate
t = 0:1/fs:1;
x = sin(2*pi*50*t) + 0.5*randn(size(t)); % 50Hz signal + noise
y = sig.filtfilt(b, a, x);

% Power spectral density
psd = sig.welch(x, fs);
figure;
plot(psd.frequencies, psd.powerSpectralDensity);
xlabel('Frequency (Hz)'); ylabel('Power Spectral Density');

statsmodels Regression

% OLS regression
n = 100;
X = randn(n, 2);
y = 1.5 + 2*X(:,1) - 3*X(:,2) + randn(n, 1)*0.5;

result = pyBridge.StatsmodelsWrapper.ols(y, X);
pyBridge.ResultParser.printResult(result);

% Logistic regression
yBinary = double(y > median(y));
logitResult = pyBridge.StatsmodelsWrapper.logistic(yBinary, X);
fprintf('AIC: %.2f, Confusion Matrix:\n', logitResult.aic);
disp(logitResult.confusionMatrix);

% Time series
y = cumsum(randn(200, 1)); % Random walk
arimaResult = pyBridge.StatsmodelsWrapper.arima(y, [1, 0, 1]);

% Unit root test
adfResult = pyBridge.StatsmodelsWrapper.adfuller(y);
fprintf('ADF test p-value: %.4f, Stationary: %s\n', ...
    adfResult.pValue, ternary(adfResult.isStationary, 'Yes', 'No'));

linearmodels Panel Data

% Panel data
nEntities = 50;
nTimes = 10;
nObs = nEntities * nTimes;

y = randn(nObs, 1);
X = randn(nObs, 2);
entityIds = repmat((1:nEntities)', nTimes, 1);
timeIds = repmat(1:nTimes, nEntities, 1);

% Fixed effects model
feResult = pyBridge.LinearmodelsWrapper.panelOLS(y, X, entityIds, timeIds);
fprintf('R-squared: %.4f\n', feResult.rSquared);

% Random effects model
reResult = pyBridge.LinearmodelsWrapper.randomEffects(y, X, entityIds, timeIds);

% Hausman test
hausman = pyBridge.LinearmodelsWrapper.hausmanTest(feResult, reResult);
fprintf('%s\n', hausman.conclusion);

Advanced Econometrics

Discrete Choice Models

%% Multinomial Logit (Multinomial Classification)
y = [0; 1; 2; 0; 1; ...]; % Multinomial dependent variable (0,1,2,...)
X = randn(500, 3);
result = pyBridge.StatsmodelsWrapper.multinomialLogit(y, X);
fprintf('Number of categories: %d\n', result.nCategories);
fprintf('Marginal effects:\n');
disp(result.marginalEffects);

%% Ordered Logit (Ordinal Choice)
yOrdered = [0; 1; 2; 3; 1; ...]; % Ordinal dependent variable (0,1,2,3,...)
% Note: addConstant=false by default (OrderedModel uses thresholds as intercepts)
result = pyBridge.StatsmodelsWrapper.orderedLogit(yOrdered, X);
fprintf('Threshold parameters:\n');
disp(result.thresholds);

%% Count Data Models
yCount = poissrnd(lambda); % Poisson-distributed count data
resultPoisson = pyBridge.StatsmodelsWrapper.poisson(yCount, X);
fprintf('Overdispersion ratio: %.3f\n', resultPoisson.overdispersionRatio);
fprintf('Overdispersion test p-value: %.4f (%s)\n', ...
    resultPoisson.overdispersionPValue, ...
    ternary(resultPoisson.hasOverdispersion, 'Significant', 'Not significant'));

% Negative binomial regression (for overdispersed data)
resultNegBin = pyBridge.StatsmodelsWrapper.negativeBinomial(yCount, X);
fprintf('Alpha parameter: %.3f\n', resultNegBin.alpha);

Robust Standard Errors

%% Heteroskedasticity-robust standard errors
resultHC0 = pyBridge.StatsmodelsWrapper.ols(y, X, covType="HC0"); % White
resultHC1 = pyBridge.StatsmodelsWrapper.ols(y, X, covType="HC1"); % Stata default
resultHC3 = pyBridge.StatsmodelsWrapper.ols(y, X, covType="HC3"); % Most conservative

%% HAC standard errors (Newey-West)
% Handle autocorrelation and heteroskedasticity
resultHAC = pyBridge.StatsmodelsWrapper.ols(y, X, ...
    covType="hac", maxLags=4, kernel="newey-west");
fprintf('HAC standard errors: [%.3f, %.3f]\n', resultHAC.stdErrors(2:3));

%% Clustered standard errors
% Single-way clustering (by firm)
resultCluster = pyBridge.StatsmodelsWrapper.ols(y, X, ...
    covType="cluster", clusterIds=firmIds);
fprintf('Number of clusters: %d\n', resultCluster.nClusters);

% Multi-way clustering (firm x year)
resultMultiway = pyBridge.StatsmodelsWrapper.ols(y, X, ...
    covType="multiway", clusterGroups={firmIds, yearIds});
fprintf('Standard errors: [%.3f, %.3f]\n', resultMultiway.stdErrors(2:3));

Panel Data + Clustered Standard Errors

%% Fixed effects + firm-clustered standard errors
result = pyBridge.LinearmodelsWrapper.panelOLS(y, X, firmIds, yearIds, ...
    entityEffects=true, covType="clustered_entity");

%% Two-way fixed effects + two-way clustered standard errors
result = pyBridge.LinearmodelsWrapper.panelOLS(y, X, firmIds, yearIds, ...
    entityEffects=true, timeEffects=true, covType="clustered_both");
fprintf('Coefficients: [%.3f, %.3f]\n', result.params);
fprintf('Standard errors: [%.3f, %.3f]\n', result.stdErrors);

econml Causal Inference

% Double Machine Learning
n = 1000;
X = randn(n, 5); % Covariates
T = 0.5*X(:,1) + randn(n, 1); % Treatment variable
Y = 2*T + 1.5*X(:,1) - 0.8*X(:,2) + randn(n, 1); % Outcome variable

dmlResult = pyBridge.EconmlWrapper.dml(Y, T, X);
fprintf('Average Treatment Effect (ATE): %.4f\n', dmlResult.ate);
fprintf('95%% Confidence Interval: [%.4f, %.4f]\n', ...
    dmlResult.ateConfInt.lower, dmlResult.ateConfInt.upper);

% Causal Forest
cfResult = pyBridge.EconmlWrapper.causalForest(Y, T, X);
fprintf('Mean CATE: %.4f\n', mean(cfResult.cate));

Documentation

Core Components

1. PyBridgeConfig - Environment Configuration

config = pyBridge.PyBridgeConfig();
config.initialize();           % Initialize Python environment
config.verifyAll();            % Verify all packages
config.printInfo();            % Print environment info
config.checkLibrary('scipy');  % Check a single library

2. DataConverter - Data Conversion

% MATLAB -> Python
pyArray = pyBridge.DataConverter.toPython(magic(3));
pyDf = pyBridge.DataConverter.table2Df(tableData);
pyDict = pyBridge.DataConverter.struct2Dict(structData);

% Python -> MATLAB
mlArray = pyBridge.DataConverter.toMatlab(pyArray);
mlTable = pyBridge.DataConverter.df2Table(pyDf);
mlStruct = pyBridge.DataConverter.dict2Struct(pyDict);

3. ResultParser - Result Parsing

% Auto-parse
result = pyBridge.ResultParser.parse(pyObject);

% Type-specific parsing
statsResult = pyBridge.ResultParser.parseStatsmodels(modelResult);
econmlResult = pyBridge.ResultParser.parseEconML(ateResult);

% Print results
pyBridge.ResultParser.printResult(result);

4. ErrorHandler - Error Handling

% Wrap calls
result = pyBridge.ErrorHandler.wrapCall(@() pyFunction());

% Manual handling
try
    result = pyFunction();
catch ME
    pyBridge.ErrorHandler.handlePyError(ME);
end

% Safe call (returns default on failure)
result = pyBridge.ErrorHandler.safeCall(@() riskyFunction(), defaultValue);

Advanced Usage

Direct Access to Python Objects

All wrappers retain access to the original Python object:

% Get original Python object
result = pyBridge.StatsmodelsWrapper.ols(y, X);
pyObj = result.originalObject;  % If original object was saved

% Call Python methods directly
pyObj.summary();

Custom Callback Functions

Use MATLAB functions in optimization and fitting:

% Custom objective function
objective = @(x) (x(1)-1)^2 + (x(2)-2.5)^2;
result = pyBridge.internal.ScipyOptimize().minimize(objective, [0, 0]);

% Custom fitting function
fitFun = @(x, a, b, c) a*exp(-b*x) + c;
result = pyBridge.internal.ScipyOptimize().curveFit(fitFun, xData, yData);

Batch Operations

% Batch hypothesis testing
stats = pyBridge.internal.ScipyStats();
tests = {'shapiro', 'normaltest', 'kstest'};
for i = 1:length(tests)
    result = stats.normalityTest(data, tests{i});
    fprintf('%s: p=%.4f\n', tests{i}, result.pValue);
end

FAQ

Q: How to change Python path?

config = pyBridge.PyBridgeConfig('path/to/python');
config.initialize();

Q: How to install missing Python packages?

% Check for missing packages
config = pyBridge.PyBridgeConfig.getInstance();
config.verifyAll();

% Install via pip (in system terminal)
% pip install scipy statsmodels linearmodels econml

Q: What data types are supported for conversion?

MATLAB -> Python:
- Numeric arrays -> NumPy arrays
- Tables -> Pandas DataFrames
- Structs -> dicts
- Cell arrays -> lists
Python -> MATLAB:
- NumPy arrays -> double arrays
- Pandas DataFrames -> tables
- dicts -> structs
- lists -> cells

Q: How to handle large datasets?

% Use chunked processing
nChunks = 10;
for i = 1:nChunks
    chunk = data((i-1)*chunkSize+1 : i*chunkSize, :);
    chunkPy = pyBridge.DataConverter.toPython(chunk);
    % Process...
end

Notes

Memory Management: Large data conversions may consume significant memory; consider chunked processing
Type Conversion: Python and MATLAB types don't map 1-to-1; watch for precision loss
Function Signatures: Refer to each library's official documentation for function parameters
Performance: Python-MATLAB interaction has overhead; for performance-critical scenarios, consider pure Python implementation

License

GPL3.0 License

References

Assistant by WorkBuddy & Qoder

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
+pyBridge		+pyBridge
+py_builtin		+py_builtin
+pyfunc		+pyfunc
docs		docs
examples		examples
outputs		outputs
tests		tests
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
startup.m		startup.m

Folders and files

Latest commit

History

Repository files navigation

PyBridge - MATLAB Python Scientific Computing Toolbox

Features

Advanced Econometrics Support

Test Status

Quick Start

1. Requirements

2. Configure Environment

3. Basic Usage

scipy Statistics

scipy Optimization

scipy Signal Processing

statsmodels Regression

linearmodels Panel Data

Advanced Econometrics

Discrete Choice Models

Robust Standard Errors

Panel Data + Clustered Standard Errors

econml Causal Inference

Documentation

Core Components

1. PyBridgeConfig - Environment Configuration

2. DataConverter - Data Conversion

3. ResultParser - Result Parsing

4. ErrorHandler - Error Handling

Advanced Usage

Direct Access to Python Objects

Custom Callback Functions

Batch Operations

FAQ

Q: How to change Python path?

Q: How to install missing Python packages?

Q: What data types are supported for conversion?

Q: How to handle large datasets?

Notes

License

References

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages