pydpf.filtering.MarginalParticleFilter

class pydpf.filtering.MarginalParticleFilter(resampler: Module | None = None, SSM: FilteringModel = None, *, use_REINFORCE_for_proposal: bool = False, use_REINFORCE_for_initial_proposal: bool = False, optimise_for_bootstrap: bool = True)

Bases: SIS

The marginal particle filter

Parameters:

resampler: Module

The resampling algorithm to use. Takes the state and log weights at the previous time-step and returns the state and log weights after resampling.

SSM: FilteringModel

A FilteringModel that represents the SSM (and optionally a proposal model). See the documentation of FilteringModel for more complete information.

use_REINFORCE_for_proposal: bool. Default False.

Whether to use the REINFORCE estimator for the gradient due to the particle proposal process.

use_REINFORCE_for_initial_proposal: bool. Default False.

Whether to use the REINFORCE estimator for the gradient due to the initial particle proposal process.

optimise_for_bootstrap: bool. Default True.

If the system is bootstrap then the filter can be optimised due to not computing terms which are guaranteed to cancel. Set to True to take advantage of these optimisations.

Notes

The marginal particle filter is a special case of the SIS algorithm. The unlike standard particle filter the marginal particle filter considers resampling in its evaluation of the proposal distribution by accounting for the possibility that any new particle could have been derived from any particle at the previous time-step instead of only considering it’s genealogical path. This is done by marginalising over the ancestral indices at each time-step, hence the name. [1]

The optimisations taken when optimise_for_bootstrap is True are only valid if the algorithm is truly bootstrap aside from the first time-step, i.e. the proposal and resampling both do not induce importance weights. However, it is permitted for SSM.initial_proposal_model to be non-bootstrap. If SSM.proposal_model is non-None then this is detected and the value of optimise_for_bootstrap is ignored. However, there is no efficient way to detect if the resampler modifies the weights so it is on the user to manually set optimise_for_bootstrap to False in this case, otherwise this filter will silently use an algorithm with undefined behaviour.

Warning

Setting the parameter optimise_for_bootstrap to True can silently invoke undefined behaviour if the resampler is non-standard, see the Notes section for detail.

References

[1]

Klaas M, de Freitas N, Doucet A (2005). “Toward Practical N 2 Monte Carlo: the Marginal Particle Filter.” In Proc. Conf. Uncert. Art. Intell. (UAI), pp. 308–315. Arlington, Virginia.

__init__(resampler: Module | None = None, SSM: FilteringModel = None, *, use_REINFORCE_for_proposal: bool = False, use_REINFORCE_for_initial_proposal: bool = False, optimise_for_bootstrap: bool = True)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Methods

__init__([resampler, SSM, ...])	Initialize internal Module state, shared by both nn.Module and ScriptModule.
add_module(name, module)	Add a child module to the current module.
apply(fn)	Apply fn recursively to every submodule (as returned by .children()) as well as self.
bfloat16()	Casts all floating point parameters and buffers to bfloat16 datatype.
buffers([recurse])	Return an iterator over module buffers.
children()	Return an iterator over immediate children modules.
compile(*args, **kwargs)	Compile this Module's forward using torch.compile().
cpu()	Move all model parameters and buffers to the CPU.
cuda([device])	Move all model parameters and buffers to the GPU.
double()	Casts all floating point parameters and buffers to double datatype.
eval()	Set the module in evaluation mode.
extra_repr()	Return the extra representation of the module.
float()	Casts all floating point parameters and buffers to float datatype.
forward(*args, **kwargs)	Run a forward pass of the SIS filter.
get_buffer(target)	Return the buffer given by target if it exists, otherwise throw an error.
get_extra_state()	Return any extra state to include in the module's state_dict.
get_parameter(target)	Return the parameter given by target if it exists, otherwise throw an error.
get_submodule(target)	Return the submodule given by target if it exists, otherwise throw an error.
half()	Casts all floating point parameters and buffers to half datatype.
ipu([device])	Move all model parameters and buffers to the IPU.
load_state_dict(state_dict[, strict, assign])	Copy parameters and buffers from state_dict into this module and its descendants.
modules([remove_duplicate])	Return an iterator over all modules in the network.
mtia([device])	Move all model parameters and buffers to the MTIA.
named_buffers([prefix, recurse, ...])	Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.
named_children()	Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.
named_modules([memo, prefix, remove_duplicate])	Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.
named_parameters([prefix, recurse, ...])	Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.
parameters([recurse])	Return an iterator over module parameters.
register_backward_hook(hook)	Register a backward hook on the module.
register_buffer(name, tensor[, persistent])	Add a buffer to the module.
register_forward_hook(hook, *[, prepend, ...])	Register a forward hook on the module.
register_forward_pre_hook(hook, *[, ...])	Register a forward pre-hook on the module.
register_full_backward_hook(hook[, prepend])	Register a backward hook on the module.
register_full_backward_pre_hook(hook[, prepend])	Register a backward pre-hook on the module.
register_load_state_dict_post_hook(hook)	Register a post-hook to be run after module's load_state_dict() is called.
register_load_state_dict_pre_hook(hook)	Register a pre-hook to be run before module's load_state_dict() is called.
register_module(name, module)	Alias for add_module().
register_parameter(name, param)	Add a parameter to the module.
register_state_dict_post_hook(hook)	Register a post-hook for the state_dict() method.
register_state_dict_pre_hook(hook)	Register a pre-hook for the state_dict() method.
requires_grad_([requires_grad])	Change if autograd should record operations on parameters in this module.
set_extra_state(state)	Set extra state contained in the loaded state_dict.
set_submodule(target, module[, strict])	Set the submodule given by target if it exists, otherwise throw an error.
share_memory()	See torch.Tensor.share_memory_().
state_dict(*args[, destination, prefix, ...])	Return a dictionary containing references to the whole state of the module.
to(*args, **kwargs)	Move and/or cast the parameters and buffers.
to_empty(*, device[, recurse])	Move the parameters and buffers to the specified device without copying storage.
train([mode])	Set the module in training mode.
type(dst_type)	Casts all parameters and buffers to dst_type.
update()	Update all constrained_parameters and cached_properties belonging to this Module.
xpu([device])	Move all model parameters and buffers to the XPU.
zero_grad([set_to_none])	Reset gradients of all model parameters.

Attributes

T_destination
call_super_init
dump_patches
training

forward(*args, **kwargs)

Run a forward pass of the SIS filter.

Parameters:

n_particles: int

The number of particles to draw per filter.

time_extent: int

The maximum time-step to run to, including time 0, the filter will draw {time_extent + 1} importance sample populations.

aggregation_function: dict[str, Module] or Module

A module that’s forward function processes the filtering outputs (the particle locations, the normalised log weights, the log sum of the unnormalised weights, the data, the time-step) into an output per time-step. Or a string indexed dictionary of such items.

observation: Tensor

The observations of the hidden variable system.

gradient_regulariser: torch.autograd.Function or None. Default None.

A autograd function to apply to the particles at the start of every time-step. It should leave the forward pass unchanged but may modify the gradients during the backward pass the intended usage is to regularise the gradient in some way. Optional.

ground_truth: Tensor or None. Default None.

The ground truth latent state. Should only be pass to the aggregation function and not used in the proposal functions. Optional.

control: Tensor or None. Default None.

The control actions. Optional.

time: Tensor or None. Default None.

The continuous time each time-step occurs at. Optional.

series_metadata: Tensor or None. Default None.

The series_metadata. Optional.

Returns:

output: Tensor or dict[str, Tensor]

The output of the filter, formed from stacking the output of aggregation_function for every time-step. Or if aggregation_function is a dictionary, a dictionary of these output Tensors one for each aggregation function.

Notes

To save memory during inference runs we allow the user to pass a function that takes a population of particles and processes this into an output for each time-step. For example, if the goal was the filtering mean then it would be wasteful to store the full population of the particles for every time-step. The memory savings are most impactful during inference as pytorch retains many intermediate tensors for gradient computation otherwise.