def __init__( self, x_ref: np.ndarray, model: Callable, p_val: float = .05, backend: Optional[str] = None, update_x_ref: Optional[Dict[str, int]] = None, uncertainty_type: str = 'dropout', n_evals: int = 25, batch_size: int = 32, device: Optional[str] = None, tokenizer: Optional[Callable] = None, max_len: Optional[int] = None, data_type: Optional[str] = None, ) -> None: """ Test for a change in the number of instances falling into regions on which the model is uncertain. Performs either a K-S test on uncertainties estimated from an preditive ensemble given either explicitly or implicitly as a model with dropout layers. Parameters ---------- x_ref Data used as reference distribution. Should be disjoint from the data the model was trained on for accurate p-values. model Regression model outputting class probabilities (or logits) backend Backend to use if model requires batch prediction. Options are 'tensorflow' or 'pytorch'. p_val p-value used for the significance of the test. update_x_ref Reference data can optionally be updated to the last n instances seen by the detector or via reservoir sampling with size n. For the former, the parameter equals {'last': n} while for reservoir sampling {'reservoir_sampling': n} is passed. uncertainty_type Method for determining the model's uncertainty for a given instance. Options are 'mc_dropout' or 'ensemble'. The former should output a scalar per instance. The latter should output a vector of predictions per instance. n_evals: The number of times to evaluate the model under different dropout configurations. Only relavent when using the 'mc_dropout' uncertainty type. batch_size Batch size used to evaluate model. Only relavent when backend has been specified for batch prediction. device Device type used. The default None tries to use the GPU and falls back on CPU if needed. Can be specified by passing either 'cuda', 'gpu' or 'cpu'. Only relevant for 'pytorch' backend. tokenizer Optional tokenizer for NLP models. max_len Optional max token length for NLP models. data_type Optionally specify the data type (tabular, image or time-series). Added to metadata. """ if backend == 'tensorflow' and not has_tensorflow or backend == 'pytorch' and not has_pytorch: raise ImportError( f'{backend} not installed. Cannot initialize and run the ' f'ClassifierUncertaintyDrift detector with {backend} backend.') if backend is None: model_fn = model else: if uncertainty_type == 'mc_dropout': if backend == 'pytorch': from alibi_detect.cd.pytorch.utils import activate_train_mode_for_dropout_layers model = activate_train_mode_for_dropout_layers(model) elif backend == 'tensorflow': logger.warning( "MC dropout being applied to tensorflow model. May not be suitable if model contains" "non-dropout layers with different train and inference time behaviour" ) from alibi_detect.cd.tensorflow.utils import activate_train_mode_for_all_layers model = activate_train_mode_for_all_layers(model) model_fn = encompass_batching(model=model, backend=backend, batch_size=batch_size, device=device, tokenizer=tokenizer, max_len=max_len) if uncertainty_type == 'mc_dropout' and backend == 'tensorflow': # To average over possible batchnorm effects as all layers evaluated in training mode. model_fn = encompass_shuffling_and_batch_filling( model_fn, batch_size=batch_size) preprocess_fn = partial(regressor_uncertainty, model_fn=model_fn, uncertainty_type=uncertainty_type, n_evals=n_evals) self._detector = KSDrift(x_ref=x_ref, p_val=p_val, preprocess_x_ref=True, update_x_ref=update_x_ref, preprocess_fn=preprocess_fn, data_type=data_type) self.meta = self._detector.meta self.meta['name'] = 'RegressorUncertaintyDrift'
def __init__( self, x_ref: np.ndarray, model: Callable, p_val: float = .05, backend: Optional[str] = None, update_x_ref: Optional[Dict[str, int]] = None, preds_type: str = 'probs', uncertainty_type: str = 'entropy', margin_width: float = 0.1, batch_size: int = 32, device: Optional[str] = None, tokenizer: Optional[Callable] = None, max_len: Optional[int] = None, data_type: Optional[str] = None, ) -> None: """ Test for a change in the number of instances falling into regions on which the model is uncertain. Performs either a K-S test on prediction entropies or Chi-squared test on 0-1 indicators of predictions falling into a margin of uncertainty (e.g. probs falling into [0.45, 0.55] in binary case). Parameters ---------- x_ref Data used as reference distribution. Should be disjoint from the data the model was trained on for accurate p-values. model Classification model outputting class probabilities (or logits) backend Backend to use if model requires batch prediction. Options are 'tensorflow' or 'pytorch'. p_val p-value used for the significance of the test. update_x_ref Reference data can optionally be updated to the last n instances seen by the detector or via reservoir sampling with size n. For the former, the parameter equals {'last': n} while for reservoir sampling {'reservoir_sampling': n} is passed. preds_type Type of prediction output by the model. Options are 'probs' (in [0,1]) or 'logits' (in [-inf,inf]). uncertainty_type Method for determining the model's uncertainty for a given instance. Options are 'entropy' or 'margin'. margin_width Width of the margin if uncertainty_type = 'margin'. The model is considered uncertain on an instance if the highest two class probabilities it assigns to the instance differ by less than margin_width. batch_size Batch size used to evaluate model. Only relavent when backend has been specified for batch prediction. device Device type used. The default None tries to use the GPU and falls back on CPU if needed. Can be specified by passing either 'cuda', 'gpu' or 'cpu'. Only relevant for 'pytorch' backend. tokenizer Optional tokenizer for NLP models. max_len Optional max token length for NLP models. data_type Optionally specify the data type (tabular, image or time-series). Added to metadata. """ if backend == 'tensorflow' and not has_tensorflow or backend == 'pytorch' and not has_pytorch: raise ImportError( f'{backend} not installed. Cannot initialize and run the ' f'ClassifierUncertaintyDrift detector with {backend} backend.') if backend is None: if device not in [None, 'cpu']: raise NotImplementedError( 'Non-pytorch/tensorflow models must run on cpu') model_fn = model else: model_fn = encompass_batching(model=model, backend=backend, batch_size=batch_size, device=device, tokenizer=tokenizer, max_len=max_len) preprocess_fn = partial( classifier_uncertainty, model_fn=model_fn, preds_type=preds_type, uncertainty_type=uncertainty_type, margin_width=margin_width, ) self._detector: Union[KSDrift, ChiSquareDrift] if uncertainty_type == 'entropy': self._detector = KSDrift(x_ref=x_ref, p_val=p_val, preprocess_x_ref=True, update_x_ref=update_x_ref, preprocess_fn=preprocess_fn, data_type=data_type) elif uncertainty_type == 'margin': self._detector = ChiSquareDrift(x_ref=x_ref, p_val=p_val, preprocess_x_ref=True, update_x_ref=update_x_ref, preprocess_fn=preprocess_fn, data_type=data_type) else: raise NotImplementedError( "Only uncertainty types 'entropy' or 'margin' supported.") self.meta = self._detector.meta self.meta['name'] = 'ClassifierUncertaintyDrift'