API reference#

Reference for simulation, fitting, parameters, and response-data utilities.

LBA model#

class pyLBA.models.linear_ballistic_accumulator.LBAParameters(A, b, v, s, tau)#

Starting-point bound, threshold, drift mean/SD, and non-decision time.

Parameters:

  • A (float | ndarray)

  • b (float | ndarray)

  • v (float | ndarray)

  • s (float | ndarray)

  • tau (float | ndarray)

validate()#

Check numeric values have compatible shapes and valid LBA support.

class pyLBA.models.linear_ballistic_accumulator.LBAModel#

An LBA race with normal drifts independently truncated above zero.

Each accumulator has a uniform starting point and constant drift per trial. The first accumulator to reach its threshold determines the response.

generate_data(n_trials, parameters, n_acc=None, seed=None)#

Simulate response times and choices with an optional random seed.

pdf(rt, response, parameters, n_acc=None)#

Return a symbolic joint density of response time and choice.

Scalars broadcast across accumulators. Vector parameters must match n_acc. Density is zero outside parameter and response-time support.

get_parameter_class()#

Return LBAParameters.

get_default_priors(n_acc, priors=None, rt_min=1.0)#

Create priors in the active PyMC model, with optional overrides.

Overrides are fixed numeric values or callables (name, n_acc) returning a PyMC variable. Defaults use positive A, b=A+gap, unit drift SD, normal drift means, and tau below the shortest observed RT.

Simulation#

pyLBA.synthetic.generate_LBA(n_trials, parameters, n_acc=None, seed=None)#

Generate synthetic response time and choice data for an accumulator model.

Parameters:

  • n_trials (int) – Number of trials to simulate.

  • parameters (LBAParameters) – Model parameters with fields A, b, v, s, tau.

  • n_acc (int, optional) – Number of accumulators.

  • seed (int, optional) – Random seed for reproducibility.

Returns:

DataFrame with columns ‘rt’ and ‘response’.

Return type:

pd.DataFrame

Core interfaces#

Import core classes from pyLBA or pyLBA.core.

class pyLBA.core.parameters.Parameter(name, value)#

A scalar or vector parameter, including symbolic PyTensor values.

class pyLBA.core.parameters.ModelParameters#

Base dataclass for named scalar or vector parameters.

to_dict()#

Return unwrapped parameter values, keyed by field name.

classmethod from_dict(values)#

Create parameters from a mapping of field names to values.

class pyLBA.core.accumulator.AccumulatorModel(name='AccumulatorModel')#

Abstract base class for accumulator models.

All accumulator models should inherit from this class and implement the required abstract methods.

Parameters:

name (str)

abstract pdf(rt, response, parameters, **kwargs)#

Compute the probability density function.

Parameters:

  • rt (np.ndarray) – Response times

  • response (np.ndarray) – Response choices (0-indexed)

  • parameters (ModelParameters) – Model parameters

Returns:

Probability density

Return type:

pt.TensorVariable

abstract get_parameter_class()#

Return the parameter class for this model.

Return type:

type

fit_mcmc(data, **kwargs)#

Fit model using MCMC.

Parameters:

  • data (pd.DataFrame) – Data with columns ‘rt’ and ‘response’

  • **kwargs – Additional arguments passed to the fitter

Returns:

Fitted model instance

Return type:

AccumulatorModel

fit_em(data, **kwargs)#

Fit model using Expectation-Maximization.

Parameters:

  • data (pd.DataFrame) – Data with columns ‘rt’ and ‘response’

  • **kwargs – Additional arguments passed to the fitter

Returns:

Fitted model instance

Return type:

AccumulatorModel

predict(n_trials, **kwargs)#

Generate predictions using posterior mean parameters.

Parameters:

  • n_trials (int) – Number of trials to simulate

  • **kwargs – Additional arguments

Returns:

Predicted data

Return type:

pd.DataFrame

property fitted: bool#

Check if model has been fitted.

property trace#

Get MCMC trace if available.

property parameters#

Get fitted parameters if available.

Fitters#

class pyLBA.fitting.MCMCFitter(model)#

MCMC fitting using PyMC.

fit(data, draws=1000, tune=1000, chains=4, cores=4, progressbar=True, priors=None, n_acc=None, **kwargs)#

Sample the posterior and return ArviZ InferenceData.

priors maps parameter names to fixed numeric values or callables (name, n_acc) creating PyMC variables inside the fitting model. n_acc defaults to the largest response label plus one. Set it explicitly when higher-numbered choices are absent from the data. Remaining keyword arguments are forwarded to pymc.sample.

class pyLBA.fitting.EMFitter(model)#

Expectation-Maximization interface.

fit(data, **kwargs)#

Raise NotImplementedError; EM fitting is not implemented.

Utilities#

Utilities for parameter broadcasting and response data.

pyLBA.utils.infer_n_acc(parameters, response=None, n_acc=None)#

Infer race size from parameter vectors and observed choice labels.

pyLBA.utils.broadcast_parameters(n, *args)#

Broadcast numeric or symbolic scalars to vectors of length n.

pyLBA.utils.validate_data(data)#

Require finite positive times and non-negative integer response labels.

pyLBA.utils.check_convergence(trace, r_hat_threshold=1.01)#

Return whether all reported R-hat values are finite and below a limit.