alibi_detect.utils.saving module

alibi_detect.utils.saving.init_ad_ae(state_dict, ae, model, model_hl)[source]

Initialize AdversarialAE.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
ae (Model) – Loaded VAE.
model (Model) – Loaded classification model.
model_hl (List[Model]) – List of tf.keras models.

Return type

AdversarialAE

Returns

Initialized AdversarialAE instance.

alibi_detect.utils.saving.init_ad_md(state_dict, distilled_model, model)[source]

Initialize ModelDistillation.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
distilled_model (Model) – Loaded distilled model.
model (Model) – Loaded classification model.

Return type

ModelDistillation

Returns

Initialized ModelDistillation instance.

alibi_detect.utils.saving.init_cd_chisquaredrift(state_dict, model, emb, tokenizer, **kwargs)[source]

Initialize ChiSquareDrift detector.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
model (Union[Model, Sequential, None]) – Optional preprocessing model.
emb (Optional[TransformerEmbedding]) – Optional text embedding model.
tokenizer (Optional[Callable]) – Optional tokenizer for text drift.
kwargs – Kwargs optionally containing preprocess_fn and preprocess_kwargs.

Return type

ChiSquareDrift

Returns

Initialized ChiSquareDrift instance.

alibi_detect.utils.saving.init_cd_classifierdrift(clf_drift, state_dict, model, emb, tokenizer, **kwargs)[source]

Initialize ClassifierDrift detector.

Parameters

clf_drift (Model) – Model used for drift classification.
state_dict (Dict) – Dictionary containing the parameter values.
model (Optional[Model]) – Optional preprocessing model.
emb (Optional[TransformerEmbedding]) – Optional text embedding model.
tokenizer (Optional[Callable]) – Optional tokenizer for text drift.
kwargs – Kwargs optionally containing preprocess_fn and preprocess_kwargs.

Return type

ClassifierDrift

Returns

Initialized ClassifierDrift instance.

alibi_detect.utils.saving.init_cd_ksdrift(state_dict, model, emb, tokenizer, **kwargs)[source]

Initialize KSDrift detector.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
model (Union[Model, Sequential, None]) – Optional preprocessing model.
emb (Optional[TransformerEmbedding]) – Optional text embedding model.
tokenizer (Optional[Callable]) – Optional tokenizer for text drift.
kwargs – Kwargs optionally containing preprocess_fn and preprocess_kwargs.

Return type

KSDrift

Returns

Initialized KSDrift instance.

alibi_detect.utils.saving.init_cd_mmddrift(state_dict, model, emb, tokenizer, **kwargs)[source]

Initialize MMDDrift detector.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
model (Union[Model, Sequential, None]) – Optional preprocessing model.
emb (Optional[TransformerEmbedding]) – Optional text embedding model.
tokenizer (Optional[Callable]) – Optional tokenizer for text drift.
kwargs – Kwargs optionally containing preprocess_fn and preprocess_kwargs.

Return type

MMDDrift

Returns

Initialized MMDDrift instance.

alibi_detect.utils.saving.init_cd_tabulardrift(state_dict, model, emb, tokenizer, **kwargs)[source]

Initialize TabularDrift detector.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
model (Union[Model, Sequential, None]) – Optional preprocessing model.
emb (Optional[TransformerEmbedding]) – Optional text embedding model.
tokenizer (Optional[Callable]) – Optional tokenizer for text drift.
kwargs – Kwargs optionally containing preprocess_fn and preprocess_kwargs.

Return type

TabularDrift

Returns

Initialized TabularDrift instance.

alibi_detect.utils.saving.init_od_ae(state_dict, ae)[source]

Initialize OutlierVAE.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
ae (Model) – Loaded AE.

Return type

OutlierAE

Returns

Initialized OutlierAE instance.

alibi_detect.utils.saving.init_od_aegmm(state_dict, aegmm)[source]

Initialize OutlierAEGMM.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
aegmm (Model) – Loaded AEGMM.

Return type

OutlierAEGMM

Returns

Initialized OutlierAEGMM instance.

alibi_detect.utils.saving.init_od_iforest(state_dict)[source]

Initialize isolation forest.

Parameters: state_dict (Dict) – Dictionary containing the parameter values.
Return type: IForest
Returns: Initialized IForest instance.

alibi_detect.utils.saving.init_od_llr(state_dict, models)[source]

Initialize LLR detector.

Parameters: state_dict (Dict) – Dictionary containing the parameter values.
Return type: LLR
Returns: Initialized LLR instance.

alibi_detect.utils.saving.init_od_mahalanobis(state_dict)[source]

Initialize Mahalanobis.

Parameters: state_dict (Dict) – Dictionary containing the parameter values.
Return type: Mahalanobis
Returns: Initialized Mahalanobis instance.

alibi_detect.utils.saving.init_od_prophet(state_dict)[source]

Initialize OutlierProphet.

Parameters: state_dict (Dict) – Dictionary containing the parameter values.
Return type: OutlierProphet
Returns: Initialized OutlierProphet instance.

alibi_detect.utils.saving.init_od_s2s(state_dict, seq2seq)[source]

Initialize OutlierSeq2Seq.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
seq2seq (Model) – Loaded seq2seq model.

Return type

OutlierSeq2Seq

Returns

Initialized OutlierSeq2Seq instance.

alibi_detect.utils.saving.init_od_sr(state_dict)[source]

Initialize spectral residual detector.

Parameters: state_dict (Dict) – Dictionary containing the parameter values.
Return type: SpectralResidual
Returns: Initialized SpectralResidual instance.

alibi_detect.utils.saving.init_od_vae(state_dict, vae)[source]

Initialize OutlierVAE.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
vae (Model) – Loaded VAE.

Return type

OutlierVAE

Returns

Initialized OutlierVAE instance.

alibi_detect.utils.saving.init_od_vaegmm(state_dict, vaegmm)[source]

Initialize OutlierVAEGMM.

Parameters

state_dict (Dict) – Dictionary containing the parameter values.
vaegmm (Model) – Loaded VAEGMM.

Return type

OutlierVAEGMM

Returns

Initialized OutlierVAEGMM instance.

alibi_detect.utils.saving.init_preprocess(state_dict, model, emb, tokenizer, **kwargs)[source]

Return preprocessing function and kwargs.

Return type: Tuple[Optional[Callable], Optional[dict]]

alibi_detect.utils.saving.load_detector(filepath, **kwargs)[source]

Load outlier, drift or adversarial detector.

Parameters: filepath (Union[str, PathLike]) – Load directory.
Return type: Union[AdversarialAE, BaseDetector, ChiSquareDrift, ClassifierDrift, ClassifierDriftTF, IForest, KSDrift, LLR, Mahalanobis, MMDDrift, MMDDriftTF, ModelDistillation, OutlierAE, OutlierAEGMM, OutlierProphet, OutlierSeq2Seq, OutlierVAE, OutlierVAEGMM, SpectralResidual, TabularDrift]
Returns: Loaded outlier or adversarial detector object.

alibi_detect.utils.saving.load_text_embed(filepath, load_dir='model')[source]

Return type: Tuple[TransformerEmbedding, Callable]

alibi_detect.utils.saving.load_tf_ae(filepath)[source]

Load AE.

Parameters: filepath (Union[str, PathLike]) – Saved model directory.
Return type: Model
Returns: Loaded AE.

alibi_detect.utils.saving.load_tf_aegmm(filepath, state_dict)[source]

Load AEGMM.

Parameters

filepath (Union[str, PathLike]) – Saved model directory.
state_dict (Dict) – Dictionary containing the n_gmm and recon_features parameters.

Return type

Model

Returns

Loaded AEGMM.

alibi_detect.utils.saving.load_tf_hl(filepath, model, state_dict)[source]

Load hidden layer models for AdversarialAE.

Parameters

filepath (Union[str, PathLike]) – Saved model directory.
model (Model) – tf.keras classification model.
state_dict (dict) – Dictionary containing the detector’s parameters.

Return type

List[Model]

Returns

List with loaded tf.keras models.

alibi_detect.utils.saving.load_tf_llr(filepath, dist_s=None, dist_b=None, input_shape=None)[source]

Load LLR TensorFlow models or distributions.

Parameters

detector – Likelihood ratio detector.
filepath (Union[str, PathLike]) – Saved model directory.
dist_s (Union[Distribution, PixelCNN, None]) – TensorFlow distribution for semantic model.
dist_b (Union[Distribution, PixelCNN, None]) – TensorFlow distribution for background model.
input_shape (Optional[tuple]) – Input shape of the model.

Returns

Detector with loaded models.

alibi_detect.utils.saving.load_tf_model(filepath, load_dir='model', custom_objects=None, model_name='model')[source]

Load TensorFlow model.

Parameters

filepath (Union[str, PathLike]) – Saved model directory.
load_dir (str) – Name of saved model folder within the filepath directory.
custom_objects (Optional[dict]) – Optional custom objects when loading the TensorFlow model.
model_name (str) – Name of loaded model.

Return type

Model

Returns

Loaded model.

alibi_detect.utils.saving.load_tf_s2s(filepath, state_dict)[source]

Load seq2seq TensorFlow model.

Parameters

filepath (Union[str, PathLike]) – Saved model directory.
state_dict (Dict) – Dictionary containing the latent_dim, shape, output_activation and beta parameters.

Return type

Model

Returns

Loaded seq2seq model.

alibi_detect.utils.saving.load_tf_vae(filepath, state_dict)[source]

Load VAE.

Parameters

filepath (Union[str, PathLike]) – Saved model directory.
state_dict (Dict) – Dictionary containing the latent dimension and beta parameters.

Return type

Model

Returns

Loaded VAE.

alibi_detect.utils.saving.load_tf_vaegmm(filepath, state_dict)[source]

Load VAEGMM.

Parameters

filepath (Union[str, PathLike]) – Saved model directory.
state_dict (Dict) – Dictionary containing the n_gmm, latent_dim and recon_features parameters.

Return type

Model

Returns

Loaded VAEGMM.

alibi_detect.utils.saving.preprocess_step_drift(cd)[source]

Return type: Tuple[Optional[Callable], Dict, Union[Model, Sequential, None], Optional[TransformerEmbedding], Dict, Optional[Callable], bool]

alibi_detect.utils.saving.save_detector(detector, filepath)[source]

Save outlier, drift or adversarial detector.

Parameters

detector (Union[AdversarialAE, BaseDetector, ChiSquareDrift, ClassifierDrift, ClassifierDriftTF, IForest, KSDrift, LLR, Mahalanobis, MMDDrift, MMDDriftTF, ModelDistillation, OutlierAE, OutlierAEGMM, OutlierProphet, OutlierSeq2Seq, OutlierVAE, OutlierVAEGMM, SpectralResidual, TabularDrift]) – Detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_embedding(embed, embed_args, filepath, save_dir='model', model_name='embedding')[source]

Save embeddings for text drift models.

Parameters

embed (Model) – Embedding model.
embed_args (dict) – Arguments for TransformerEmbedding module.
filepath (Union[str, PathLike]) – Save directory.
save_dir (str) – Name of folder to save to within the filepath directory.
model_name (str) – Name of saved model.

Return type

None

alibi_detect.utils.saving.save_tf_ae(detector, filepath)[source]

Save TensorFlow components of OutlierAE

Parameters

detector (Union[OutlierAE, AdversarialAE]) – Outlier or adversarial detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_tf_aegmm(od, filepath)[source]

Save TensorFlow components of OutlierAEGMM.

Parameters

od (OutlierAEGMM) – Outlier detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_tf_hl(models, filepath)[source]

Save TensorFlow model weights.

Parameters

models (List[Model]) – List with tf.keras models.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_tf_llr(detector, filepath)[source]

Save LLR TensorFlow models or distributions.

Parameters

detector (LLR) – Outlier detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_tf_model(model, filepath, save_dir='model', model_name='model')[source]

Save TensorFlow model.

Parameters

model (Model) – tf.keras.Model or tf.keras.Sequential.
filepath (Union[str, PathLike]) – Save directory.
save_dir (str) – Name of folder to save to within the filepath directory.
model_name (str) – Name of saved model.

Return type

None

alibi_detect.utils.saving.save_tf_s2s(od, filepath)[source]

Save TensorFlow components of OutlierSeq2Seq.

Parameters

od (OutlierSeq2Seq) – Outlier detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_tf_vae(detector, filepath)[source]

Save TensorFlow components of OutlierVAE.

Parameters

detector (OutlierVAE) – Outlier detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.save_tf_vaegmm(od, filepath)[source]

Save TensorFlow components of OutlierVAEGMM.

Parameters

od (OutlierVAEGMM) – Outlier detector object.
filepath (Union[str, PathLike]) – Save directory.

Return type

None

alibi_detect.utils.saving.state_adv_ae(ad)[source]

AdversarialAE parameters to save.

Parameters: ad (AdversarialAE) – Adversarial detector object.
Return type: Dict

alibi_detect.utils.saving.state_adv_md(md)[source]

ModelDistillation parameters to save.

Parameters: md (ModelDistillation) – ModelDistillation detector object.
Return type: Dict

alibi_detect.utils.saving.state_ae(od)[source]

OutlierAE parameters to save.

Parameters: od (OutlierAE) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_aegmm(od)[source]

OutlierAEGMM parameters to save.

Parameters: od (OutlierAEGMM) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_chisquaredrift(cd)[source]

Chi-Squared drift detector parameters to save.

Parameters: cd (ChiSquareDrift) – Drift detection object.
Return type: Tuple[Dict, Union[Model, Sequential, None], Optional[TransformerEmbedding], Optional[Dict], Optional[Callable]]

alibi_detect.utils.saving.state_classifierdrift(cd)[source]

Classifier-based drift detector parameters to save.

Parameters: cd (ClassifierDrift) – Drift detection object.
Return type: Tuple[Dict, Union[Sequential, Model], Union[Model, Sequential, None], Optional[TransformerEmbedding], Optional[Dict], Optional[Callable]]

alibi_detect.utils.saving.state_iforest(od)[source]

Isolation forest parameters to save.

Parameters: od (IForest) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_ksdrift(cd)[source]

K-S drift detector parameters to save.

Parameters: cd (KSDrift) – Drift detection object.
Return type: Tuple[Dict, Union[Model, Sequential, None], Optional[TransformerEmbedding], Optional[Dict], Optional[Callable]]

alibi_detect.utils.saving.state_llr(od)[source]

LLR parameters to save.

Parameters: od (LLR) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_mahalanobis(od)[source]

Mahalanobis parameters to save.

Parameters: od (Mahalanobis) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_mmddrift(cd)[source]

MMD drift detector parameters to save. Note: only GaussianRBF kernel supported.

Parameters: cd (MMDDrift) – Drift detection object.
Return type: Tuple[Dict, Union[Model, Sequential, None], Optional[TransformerEmbedding], Optional[Dict], Optional[Callable]]

alibi_detect.utils.saving.state_prophet(od)[source]

OutlierProphet parameters to save.

Parameters: od (OutlierProphet) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_s2s(od)[source]

OutlierSeq2Seq parameters to save.

Parameters: od (OutlierSeq2Seq) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_sr(od)[source]

Spectral residual parameters to save.

Parameters: od (SpectralResidual) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_tabulardrift(cd)[source]

Tabular drift detector parameters to save.

Parameters: cd (TabularDrift) – Drift detection object.
Return type: Tuple[Dict, Union[Model, Sequential, None], Optional[TransformerEmbedding], Optional[Dict], Optional[Callable]]

alibi_detect.utils.saving.state_vae(od)[source]

OutlierVAE parameters to save.

Parameters: od (OutlierVAE) – Outlier detector object.
Return type: Dict

alibi_detect.utils.saving.state_vaegmm(od)[source]

OutlierVAEGMM parameters to save.

Parameters: od (OutlierVAEGMM) – Outlier detector object.
Return type: Dict