evofr.infer package

Submodules

evofr.infer.BJBackendsScrap module

evofr.infer.InferMCMC module

class InferMCMC(num_warmup, num_samples, kernel, **kernel_kwargs)

Bases: object

Parameters:

  • num_warmup (int)

  • num_samples (int)

  • kernel (Type[MCMCKernel])

fit(model, data, name=None)

Fit model given data using specificed MCMC method.

Parameters:

  • model (ModelSpec) – ModelSpec for model

  • data (DataSpec) – DataSpec for data to do inference on

  • name (str | None) – name used to index posterior

Return type:

PosteriorHandler

class InferNUTS(num_warmup, num_samples, **kernel_kwargs)

Bases: InferMCMC

Parameters:

  • num_warmup (int)

  • num_samples (int)

class InferNUTS_from_MAP(num_warmup, num_samples, iters, lr)

Bases: object

fit(model, data, name=None)

evofr.infer.InferSVI module

class InferFullRank(iters, lr, num_samples, **handler_kwargs)

Bases: InferSVI

Parameters:

  • iters (int)

  • lr (float)

  • num_samples (int)

class InferMAP(iters, lr, **handler_kwargs)

Bases: InferSVI

Parameters:

  • iters (int)

  • lr (float)

class InferSVI(iters, lr, num_samples, guide_fn, **handler_kwargs)

Bases: object

Parameters:

  • iters (int)

  • lr (float)

  • num_samples (int)

  • guide_fn (Type[AutoGuide])

fit(model, data, name=None)

Fit model given data using specificed SVI method.

Parameters:

  • model (ModelSpec) – ModelSpec for model

  • data (DataSpec) – DataSpec for data to do inference on

  • name (str | None) – name used to index posterior

Return type:

PosteriorHandler

init_to_MAP(model, data, iters=10000, lr=0.004)

Initilization strategy for MCMC. Estimates MAP for given model and data. Returns initilization strategy and MAP estimates.

Parameters:

evofr.infer.MCMC_handler module

class MCMCHandler(rng_key=None, kernel=None, **kernel_kwargs)

Bases: object

Parameters:

  • rng_key (Array | None)

  • kernel (Type[MCMCKernel] | None)

fit(model, data, num_warmup, num_samples, **mcmc_kwargs)

Fit model using MCMC given data.

model:

a numpyro model.

data:

dictionary containing arguments to ‘model’.

num_warmup:

number of samples for warmup period in MCMC.

num_samples:

number of samples to be returned in MCMC.

mcmc_kwargs:

additional arguments to be passed to MCMC algorithms.

Parameters:

  • model (Callable)

  • data (Dict)

  • num_warmup (int)

  • num_samples (int)

load_state(file_path)
property params: Dict
predict(model, data, **kwargs)
Parameters:

  • model (Callable)

  • data (Dict)

save_state(file_path)

evofr.infer.SVI_handler module

class SVIHandler(rng_key=None, loss=None, optimizer=None)

Bases: object

Parameters:

  • rng_key (Array | None)

  • loss (Trace_ELBO | None)

  • optimizer (Any | None)

fit(model, guide, data, n_epochs)
Parameters:

  • model (Callable)

  • guide (AutoGuide)

  • data (dict)

  • n_epochs (int)

init_svi(model, guide, data)
Parameters:

  • model (Callable)

  • guide (AutoGuide)

  • data (dict)

load_state(fp)
property losses
property optim_state
property params
predict(model, guide, data, **kwargs)
reset_state()
save_state(fp)

Module contents