mmd

The Maximum Mean Discrepancy (MMD) approach uses the distance measure of the same name to adapt a feature extractor. This implementation uses a multi-kernel variant of the MMD loss with bandwidths set via the median heuristic.

Source --> FeatEx --> Source Feats -----------> Regressor  --> RUL Prediction
        ^         |                 |
        |         |                 v
Target --         --> Target Feats -->  MMD Loss

It was first introduced by Long et al. as Deep Adaption Network (DAN) for image classification.

Used In

• Cao et al. (2021). Transfer learning for remaining useful life prediction of multi-conditions bearings based on bidirectional-GRU network. Measurement: Journal of the International Measurement Confederation, 178. 10.1016/j.measurement.2021.109287
• Krokotsch et al. (2020). A Novel Evaluation Framework for Unsupervised Domain Adaption on Remaining Useful Lifetime Estimation. 2020 IEEE International Conference on Prognostics and Health Management (ICPHM). 10.1109/ICPHM49022.2020.9187058

MmdApproach

Bases: AdaptionApproach

The MMD uses the Maximum Mean Discrepancy to adapt a feature extractor to be used with the source regressor.

The regressor needs the same number of input units as the feature extractor has output units.

Examples:

>>> from rul_adapt import model
>>> from rul_adapt import approach
>>> feat_ex = model.CnnExtractor(1, [16, 16, 1], 10, fc_units=16)
>>> reg = model.FullyConnectedHead(16, [1])
>>> mmd = approach.MmdApproach(0.01)
>>> mmd.set_model(feat_ex, reg)

__init__(mmd_factor, num_mmd_kernels=5, loss_type='mse', rul_score_mode='phm08', evaluate_degraded_only=False, **optim_kwargs)

Create a new MMD approach.

The strength of the influence of the MMD loss on the feature extractor is controlled by the mmd_factor. The higher it is, the stronger the influence.

For more information about the possible optimizer keyword arguments, see here.

Parameters:

Name	Type	Description	Default
mmd_factor	float	The strength of the MMD loss' influence.	required
num_mmd_kernels	int	The number of kernels for the MMD loss.	5
loss_type	Literal['mse', 'rmse', 'mae']	The type of regression loss, either 'mse', 'rmse' or 'mae'.	'mse'
rul_score_mode	Literal['phm08', 'phm12']	The mode for the val and test RUL score, either 'phm08' or 'phm12'.	'phm08'
evaluate_degraded_only	bool	Whether to only evaluate the RUL score on degraded samples.	False
**optim_kwargs	Any	Keyword arguments for the optimizer, e.g. learning rate.	{}

configure_optimizers()

Configure an optimizer.

forward(inputs)

Predict the RUL values for a batch of input features.

test_step(batch, batch_idx, dataloader_idx)

Execute one test step.

The batch argument is a list of two tensors representing features and labels. A RUL prediction is made from the features and the validation RMSE and RUL score are calculated. The metrics recorded for dataloader idx zero are assumed to be from the source domain and for dataloader idx one from the target domain. The metrics are written to the configured logger under the prefix test.

Parameters:

Name	Type	Description	Default
batch	List[Tensor]	A list containing a feature and a label tensor.	required
batch_idx	int	The index of the current batch.	required
dataloader_idx	int	The index of the current dataloader (0: source, 1: target).	required

training_step(batch, batch_idx)

Execute one training step.

The batch argument is a list of three tensors representing the source features, source labels and target features. Both types of features are fed to the feature extractor. Then the regression loss for the source domain and the MMD loss between domains is computed. The regression, MMD and combined loss are logged.

Parameters:

Name	Type	Description	Default
batch	List[Tensor]	A list of a source feature, source label and target feature tensors.	required
batch_idx	int	The index of the current batch.	required

Returns: The combined loss.

validation_step(batch, batch_idx, dataloader_idx)

Execute one validation step.

The batch argument is a list of two tensors representing features and labels. A RUL prediction is made from the features and the validation RMSE and RUL score are calculated. The metrics recorded for dataloader idx zero are assumed to be from the source domain and for dataloader idx one from the target domain. The metrics are written to the configured logger under the prefix val.

Parameters:

Name	Type	Description	Default
batch	List[Tensor]	A list containing a feature and a label tensor.	required
batch_idx	int	The index of the current batch.	required
dataloader_idx	int	The index of the current dataloader (0: source, 1: target).	required