timesead.models.baselines

Submodules

Classes

WMDAnomalyDetector

Base class for all neural network modules.

Base_ThresholdAD

Base class for all neural network modules.

OOSAnomalyDetector

Base class for all neural network modules.

IQRAnomalyDetector

Base class for all neural network modules.

PCAAnomalyDetector

Base class for all neural network modules.

Package Contents

class timesead.models.baselines.WMDAnomalyDetector(first_diffs: bool = False, cum_method: str = 'max', feature_index: int | None = None, full_cov: bool = False, input_shape: str = 'btf')

Bases: timesead.models.common.AnomalyDetector

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing them to be nested in a tree structure. You can assign the submodules as regular attributes:

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will also have their parameters converted when you call to(), etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

Variables:

training (bool) – Boolean represents whether this module is in training or evaluation mode.

Parameters:

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

  • full_cov (bool)

  • input_shape (str)

Weighted Mean Distance Anomaly Detector

A simlpe anomaly detector that outputs the distance to the mean (mean from training data) weighted by each feature standard deviation.

Parameters:

  • first_diffs[bool] – Flag, if instead of raw values first difference should be used. Default is False.

  • cum_method[str] – One of {‘mean’, ‘max’}. Accumulation method over feature dimension. Note that when feature_index is not None. Then accumulation method will be ignored. Default is ‘max’.

  • feature_index[optional[int]] – Take scores for specific feature. If none, above accumulation rule will be used. Default is None.

  • full_cov[bool] – Take full covariance matrix to weight diviation from the mean. Default is False.

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

  • full_cov (bool)

  • input_shape (str)

first_diffs = False
cum_method = ''
feature = None
full_cov = False
input_shape = 'btf'
mean = None
inv_std = None
fit(dataset: torch.utils.data.DataLoader) → None

Fit this anomaly detector on a dataset. Note that we assume only normal data here.

Parameters:

dataset (torch.utils.data.DataLoader) – A dataset

Return type:

None

compute_online_anomaly_score(inputs: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Compute the online anomaly score for a batch of inputs. The output tensor must have the same shape as the output of format_targets when called with the corresponding targets for this batch. This method expects a window (or a batch of windows) as its input and should return a score for the last point in the window.

Parameters:

inputs (Tuple[torch.Tensor, Ellipsis]) – tuple of input tensors

Returns:

Tensor of shape (B,) that contains the anomaly scores for this batch

Return type:

torch.Tensor

abstract compute_offline_anomaly_score(inputs: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Compute the offline anomaly score for a batch of inputs. The output tensor must have the same shape as the output of format_targets when called with the corresponding targets for this batch. This method expects a window (or a batch of windows) as its input and should return a score for the last point in the window.

Parameters:

inputs (Tuple[torch.Tensor, Ellipsis]) – tuple of input tensors

Returns:

Tensor of shape (N,) that contains the anomaly scores for this batch

Return type:

torch.Tensor

format_online_targets(targets: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Format the labels for a batch of targets. The output tensor must have the same shape as the output of compute_online_anomaly_score when called with the corresponding inputs for this batch.

Parameters:

targets (Tuple[torch.Tensor, Ellipsis]) – tuple of target tensors

Returns:

Tensor of shape (B,) that contains the ground truth labels for this batch

Return type:

torch.Tensor

class timesead.models.baselines.Base_ThresholdAD(first_diffs: bool, cum_method: str, feature_index: int | None, input_shape: str = 'btf')

Bases: timesead.models.common.AnomalyDetector

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing them to be nested in a tree structure. You can assign the submodules as regular attributes:

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will also have their parameters converted when you call to(), etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

Variables:

training (bool) – Boolean represents whether this module is in training or evaluation mode.

Parameters:

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

  • input_shape (str)

A Basis Threshold Anomaly Detector

A simlpe anomaly detector that is equals zero for all data within the given Threshold and else equals the distance to the given lower/upper thr of training value.

Parameters:

  • first_diffs[bool] – Flag, if instead of raw values first difference should be used.

  • cum_method[str] – One of {‘mean’, ‘max’}. Accumulation method over feature dimension. Note that when feature_index is not None. Then accumulation method will be ignored.

  • feature_index[optional[int]] – Take scores for specific feature. If none, above accumulation rule will be used.

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

  • input_shape (str)

first_diffs
cum_method
feature
input_shape = 'btf'
lower_thresh = None
upper_thresh = None
compute_online_anomaly_score(inputs: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Compute the online anomaly score for a batch of inputs. The output tensor must have the same shape as the output of format_targets when called with the corresponding targets for this batch. This method expects a window (or a batch of windows) as its input and should return a score for the last point in the window.

Parameters:

inputs (Tuple[torch.Tensor, Ellipsis]) – tuple of input tensors

Returns:

Tensor of shape (B,) that contains the anomaly scores for this batch

Return type:

torch.Tensor

abstract compute_offline_anomaly_score(inputs: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Compute the offline anomaly score for a batch of inputs. The output tensor must have the same shape as the output of format_targets when called with the corresponding targets for this batch. This method expects a window (or a batch of windows) as its input and should return a score for the last point in the window.

Parameters:

inputs (Tuple[torch.Tensor, Ellipsis]) – tuple of input tensors

Returns:

Tensor of shape (N,) that contains the anomaly scores for this batch

Return type:

torch.Tensor

format_online_targets(targets: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Format the labels for a batch of targets. The output tensor must have the same shape as the output of compute_online_anomaly_score when called with the corresponding inputs for this batch.

Parameters:

targets (Tuple[torch.Tensor, Ellipsis]) – tuple of target tensors

Returns:

Tensor of shape (B,) that contains the ground truth labels for this batch

Return type:

torch.Tensor

class timesead.models.baselines.OOSAnomalyDetector(first_diffs: bool = False, cum_method: str = 'mean', feature_index: int | None = None, *args, **kwargs)

Bases: Base_ThresholdAD

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing them to be nested in a tree structure. You can assign the submodules as regular attributes:

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will also have their parameters converted when you call to(), etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

Variables:

training (bool) – Boolean represents whether this module is in training or evaluation mode.

Parameters:

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

Out of Support Anomaly Detector

A simlpe anomaly detector that is equals zero for all data within the data support of the training data and else equals the distance to the min/max of training value.

Parameters:

  • first_diffs[bool] – Flag, if instead of raw values first difference should be used. Default is False.

  • cum_method[str] – One of {‘mean’, ‘max’}. Accumulation method over feature dimension. Note that when feature_index is not None. Then accumulation method will be ignored. Default is ‘mean’.

  • feature_index[optional[int]] – Take scores for specific feature. If none, above accumulation rule will be used. Default is None.

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

fit(dataset: torch.utils.data.DataLoader) → None

Fit this anomaly detector on a dataset. Note that we assume only normal data here.

Parameters:

dataset (torch.utils.data.DataLoader) – A dataset

Return type:

None

class timesead.models.baselines.IQRAnomalyDetector(std_factor: float = 2.58, first_diffs: bool = False, cum_method: str = 'mean', feature_index: int | None = None, *args, **kwargs)

Bases: Base_ThresholdAD

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing them to be nested in a tree structure. You can assign the submodules as regular attributes:

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will also have their parameters converted when you call to(), etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

Variables:

training (bool) – Boolean represents whether this module is in training or evaluation mode.

Parameters:

  • std_factor (float)

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

Interquantile Range Anomaly Detector

A simlpe anomaly detector that is equals zero for all data within some interquantile range of the normal training data and else equals the distance to nearest quantile border.

Parameters:

  • std_factor[float] – float, that gives the width of the IQR. Default is 2.58 (=99.5% normal quantile).

  • first_diffs[bool] – Flag, if instead of raw values first difference should be used. Default is False.

  • cum_method[str] – One of {‘mean’, ‘max’}. Accumulation method over feature dimension. Note that when feature_index is not None. Then accumulation method will be ignored. Default is ‘mean’.

  • feature_index[optional[int]] – Take scores for specific feature. If none, above accumulation rule will be used. Default is None.

  • std_factor (float)

  • first_diffs (bool)

  • cum_method (str)

  • feature_index (Optional[int])

std_factor = 2.58
fit(dataset: torch.utils.data.DataLoader) → None

Fit this anomaly detector on a dataset. Note that we assume only normal data here.

Parameters:

dataset (torch.utils.data.DataLoader) – A dataset

Return type:

None

class timesead.models.baselines.PCAAnomalyDetector(n_components: int | float, pca_method: str = 'standard', first_diffs: bool = False, cum_method: str = 'mean', input_shape: str = 'btf', **kwargs)

Bases: timesead.models.common.AnomalyDetector

Base class for all neural network modules.

Your models should also subclass this class.

Modules can also contain other Modules, allowing them to be nested in a tree structure. You can assign the submodules as regular attributes:

import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

Submodules assigned in this way will be registered, and will also have their parameters converted when you call to(), etc.

Note

As per the example above, an __init__() call to the parent class must be made before assignment on the child.

Variables:

training (bool) – Boolean represents whether this module is in training or evaluation mode.

Parameters:

  • n_components (Union[int, float])

  • pca_method (str)

  • first_diffs (bool)

  • cum_method (str)

  • input_shape (str)

PCA Anomaly Detector

A simlpe anomaly detector that uses PCA as reconstruction method.

Parameters:

  • n_components[int – number of principle components to keep. If float with 0<n_comp<=1.,the number of components will equal the explained variance, note, that this only works for standard pca. Defaults equals 0.95.

  • float] – number of principle components to keep. If float with 0<n_comp<=1.,the number of components will equal the explained variance, note, that this only works for standard pca. Defaults equals 0.95.

  • pca_method[str – {‘standard’,’kernel’}]: Which PCA method should be used. Default is standard.

  • first_diffs[bool] – Flag, if instead of raw values first difference should be used. Default is False.

  • cum_method[str – {‘mean’,’max’}]: Accumulation method over feature dimension. Note that when feature_index is not None. Then accumulation method will be ignored. Default is ‘mean’.

  • n_components (Union[int, float])

  • pca_method (str)

  • first_diffs (bool)

  • cum_method (str)

  • input_shape (str)

first_diffs = False
cum_method = ''
input_shape = 'btf'
fit(dataset: torch.utils.data.DataLoader) → None

Fit this anomaly detector on a dataset. Note that we assume only normal data here.

Parameters:

dataset (torch.utils.data.DataLoader) – A dataset

Return type:

None

compute_online_anomaly_score(inputs: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Compute the online anomaly score for a batch of inputs. The output tensor must have the same shape as the output of format_targets when called with the corresponding targets for this batch. This method expects a window (or a batch of windows) as its input and should return a score for the last point in the window.

Parameters:

inputs (Tuple[torch.Tensor, Ellipsis]) – tuple of input tensors

Returns:

Tensor of shape (B,) that contains the anomaly scores for this batch

Return type:

torch.Tensor

abstract compute_offline_anomaly_score(inputs: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Compute the offline anomaly score for a batch of inputs. The output tensor must have the same shape as the output of format_targets when called with the corresponding targets for this batch. This method expects a window (or a batch of windows) as its input and should return a score for the last point in the window.

Parameters:

inputs (Tuple[torch.Tensor, Ellipsis]) – tuple of input tensors

Returns:

Tensor of shape (N,) that contains the anomaly scores for this batch

Return type:

torch.Tensor

format_online_targets(targets: Tuple[torch.Tensor, Ellipsis]) → torch.Tensor

Format the labels for a batch of targets. The output tensor must have the same shape as the output of compute_online_anomaly_score when called with the corresponding inputs for this batch.

Parameters:

targets (Tuple[torch.Tensor, Ellipsis]) – tuple of target tensors

Returns:

Tensor of shape (B,) that contains the ground truth labels for this batch

Return type:

torch.Tensor