pyprobound.fitting.BaseFit

class BaseFit(rnd, dataset, prediction, observation=<function BaseFit.<lambda>>, update_construct=False, train_posbias=False, train_hill=False, max_split=None, batch_size=None, checkpoint='valmodel.pt', output='/dev/null', device=None, sampler=None, optimizer=<class 'torch.optim.lbfgs.LBFGS'>, optim_args=None, sampler_args=None, name='')

Bases: BaseLoss[CountBatch], ABC

Base class for curve fitting to independent validation data.

__init__(rnd, dataset, prediction, observation=<function BaseFit.<lambda>>, update_construct=False, train_posbias=False, train_hill=False, max_split=None, batch_size=None, checkpoint='valmodel.pt', output='/dev/null', device=None, sampler=None, optimizer=<class 'torch.optim.lbfgs.LBFGS'>, optim_args=None, sampler_args=None, name='')

Initializes the curve fitting.

Parameters:

• rnd (BaseRound | Aggregate) – A component containing an aggregate of different modes.

• dataset (CountTable) – A CountTable with 1 to 3 columns, with the first column taken as the target; if 2 columns are provided, the second column is taken as a symmetrical error; if 3 columns are provided, the second is taken as the lower error and the third is taken as the upper error.

• prediction (Callable[[Tensor], Tensor]) – A callable applied to the log aggregate \(\log Z\).

• observation (Callable[[Tensor], Tensor]) – A callable applied to the target \(y\).

• update_construct (bool) – Whether to reset experiment-specific parameters.

• train_posbias (bool) – Whether to retrain positional bias profiles \(\omega\).

• train_hill (bool) – Whether to train a Hill coefficient.

• max_split (int | None) – Maximum number of sequences scored at a time (lower values reduce memory but increase computation time).

• batch_size (int | None) – The number of sequences used to optimize the model at a time.

• checkpoint (str | PathLike[str]) – The file where the model will be checkpointed to.

• output (str | PathLike[str]) – The file where the optimization output will be written to.

• device (str | None) – The device on which to perform optimization.

• sampler (type[Sampler[CountBatch]] | None) – The sampler used when creating the dataloader.

• optimizer (type[Optimizer]) – The optimizer used for optimization.

• optim_args (MutableMapping[str, Any] | None) – Parameters passed to the optimizer. (Defaults to {“line_search_fn”:”strong_wolfe”} if available).

• sampler_args (MutableMapping[str, Any] | None) – Parameters passed to the sampler.

• name (str) – A string used to describe the validation dataset.

Methods

check_length_consistency()	Checks that input lengths of Binding components are consistent.
components()	Iterator of child components.
forward(batches)	Calculates the multitask weighted loss and regularization.
freeze()	Turns off gradient calculation for all parameters.
get_setup_string()	A description used when printing the output of an optimizer.
log_aggregate(seqs)	Calculates the log aggregate \(\log Z_i\).
max_embedding_size()	The maximum number of bytes needed to encode a sequence.
negloglik(transform, batch)	Calculates the negative log-likelihood plus a normalization factor.
obs_pred(seqs, target)	Calculates the observed and predicted values used for the loss.
optim_procedure([ancestry, current_order])	The sequential optimization procedure for all Binding components.
regularization(component)	Calculates parameter regularization.
reload(checkpoint)	Loads the model from a checkpoint file.
reload_from_state_dict(state_dict)	Loads the model from a state dict.
save(checkpoint[, flank_lengths])	Saves the model to a file with "state_dict" and "metadata" fields.
score(batch)	Wraps obs_pred, automatically managing devices.
unfreeze([parameter])	Turns on gradient calculation for the specified parameter.

Attributes

unfreezable	alias of Literal['all']
scale
intercept

Non-Inherited Members

scale: Parameter

intercept: Parameter

get_setup_string()

A description used when printing the output of an optimizer.

Return type:

str

log_aggregate(seqs)

Calculates the log aggregate \(\log Z_i\).

Parameters:

seqs (Tensor) – A sequence tensor of shape \((\text{minibatch},\text{length})\) or \((\text{minibatch},\text{in_channels},\text{length})\).

Return type:

Tensor

Returns:

The log aggregate tensor of shape \((\text{minibatch},)\).

abstract obs_pred(seqs, target)

Calculates the observed and predicted values used for the loss.

Parameters:

• seqs (Tensor) – A sequence tensor of shape \((\text{minibatch},\text{length})\) or \((\text{minibatch},\text{in_channels},\text{length})\).

• target (Tensor) – A target tensor of shape \((\text{minibatch},1-3)\)

Return type:

tuple[Tensor, Tensor, Tensor | None, Tensor | None]

Returns:

A tuple of four tensors of shape \((\text{minibatch},)\), being the transformed observed values, the transformed predicted values, the lower error values, and the upper error values.

score(batch)

Wraps obs_pred, automatically managing devices.

Return type:

tuple[Tensor, Tensor, Tensor | None, Tensor | None]

negloglik(transform, batch)

Calculates the negative log-likelihood plus a normalization factor.

Parameters:

• transform (Transform) – The component used for scoring.

• batch (CountBatch) – The batch to be scored.

Return type:

tuple[Tensor, int]

Returns:

A tuple of scalar tensors (negloglik, norm), where negloglik is the negative log-likelihood of the batch and norm is a scaling factor. A running sum of each is kept, and the loss is the ratio.