def predict(self, X: Union[np.ndarray, list], drift_type: str = 'batch',
return_p_val: bool = True, return_distance: bool = True) \
-> Dict[Dict[str, str], Dict[str, Union[np.ndarray, int, float]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
X
Batch of instances.
drift_type
Predict drift at the 'feature' or 'batch' level. For 'batch', the K-S statistics for
each feature are aggregated using the Bonferroni or False Discovery Rate correction.
return_p_val
Whether to return feature level p-values.
return_distance
Whether to return the K-S statistic between the features of the new batch and reference data.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift predictions and both feature and batch level drift scores.
'data' contains the drift prediction and optionally the feature level p-values,
threshold after multivariate correction if needed and K-S statistics.
"""
# compute drift scores
p_vals, dist = self.score(X)
# values below p-value threshold are drift
if drift_type == 'feature':
drift_pred = (p_vals < self.p_val).astype(int)
elif drift_type == 'batch' and self.correction == 'bonferroni':
threshold = self.p_val / self.n_features
drift_pred = int((p_vals < threshold).any())
elif drift_type == 'batch' and self.correction == 'fdr':
drift_pred, threshold = fdr(p_vals, q_val=self.p_val)
else:
raise ValueError(
'`drift_type` needs to be either `feature` or `batch`.')
# update reference dataset
if (isinstance(self.update_X_ref, dict)
and self.preprocess_fn is not None and self.preprocess_X_ref):
X = self.preprocess_fn(X)
self.X_ref = update_reference(self.X_ref, X, self.n, self.update_X_ref)
# used for reservoir sampling
self.n += X.shape[0] # type: ignore
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_vals
cd['data'][
'threshold'] = self.p_val if drift_type == 'feature' else threshold
if return_distance:
cd['data']['distance'] = dist
return cd
def predict(self, x: Union[np.ndarray, list], return_p_val: bool = True,
return_distance: bool = True, return_probs: bool = True, return_model: bool = True) \
-> Dict[str, Dict[str, Union[str, int, float, Callable]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
x
Batch of instances.
return_p_val
Whether to return the p-value of the test.
return_distance
Whether to return a notion of strength of the drift.
K-S test stat if binarize_preds=False, otherwise relative error reduction.
return_probs
Whether to return the instance level classifier probabilities for the reference and test data
(0=reference data, 1=test data).
return_model
Whether to return the updated model trained to discriminate reference and test instances.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the p-value, performance of the classifier
relative to its expectation under the no-change null, the out-of-fold classifier model
prediction probabilities on the reference and test data, and the trained model.
"""
# compute drift scores
p_val, dist, probs_ref, probs_test = self.score(x)
drift_pred = int(p_val < self.p_val)
# update reference dataset
if isinstance(
self.update_x_ref, dict
) and self.preprocess_fn is not None and self.preprocess_x_ref:
x = self.preprocess_fn(x)
# TODO: TBD: can `x` ever be a `list` after pre-processing? update_references and downstream functions
# don't support list inputs and without the type: ignore[arg-type] mypy complains
self.x_ref = update_reference(
self.x_ref, x, self.n, self.update_x_ref) # type: ignore[arg-type]
# used for reservoir sampling
self.n += len(x)
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_val
cd['data']['threshold'] = self.p_val
if return_distance:
cd['data']['distance'] = dist
if return_probs:
cd['data']['probs_ref'] = probs_ref
cd['data']['probs_test'] = probs_test
if return_model:
cd['data']['model'] = self.model
return cd
def predict(self, # type: ignore[override]
x: Union[np.ndarray, list], c: np.ndarray,
return_p_val: bool = True, return_distance: bool = True, return_coupling: bool = False) \
-> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether a batch of data has drifted from the reference data, given the provided context.
Parameters
----------
x
Batch of instances.
c
Context associated with batch of instances.
return_p_val
Whether to return the p-value of the permutation test.
return_distance
Whether to return the conditional MMD test statistic between the new batch and reference data.
return_coupling
Whether to return the coupling matrices.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the p-value, threshold, conditional MMD test statistic
and coupling matrices.
"""
# compute drift scores
p_val, dist, distance_threshold, coupling = self.score(x, c)
drift_pred = int(p_val < self.p_val)
# update reference dataset
if isinstance(
self.update_ref, dict
) and self.preprocess_fn is not None and self.preprocess_x_ref:
x = self.preprocess_fn(x)
self.x_ref = update_reference(
self.x_ref, x, self.n, self.update_ref) # type: ignore[arg-type]
self.c_ref = update_reference(
self.c_ref, c, self.n, self.update_ref) # type: ignore[arg-type]
# used for reservoir sampling
self.n += len(x)
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_val
cd['data']['threshold'] = self.p_val
if return_distance:
cd['data']['distance'] = dist
cd['data']['distance_threshold'] = distance_threshold
if return_coupling:
cd['data']['coupling_xx'] = coupling[0]
cd['data']['coupling_yy'] = coupling[1]
cd['data']['coupling_xy'] = coupling[2]
return cd
def predict(
self,
X: np.ndarray,
drift_type: str = "batch",
return_p_val: bool = True
) -> Dict[Dict[str, str], Dict[str, np.ndarray]]:
cd = concept_drift_dict()
cd["data"]["is_drift"] = self.expect_return_is_drift
return cd
def predict(
self,
X: np.ndarray,
drift_type: str = "batch",
return_p_val: bool = True
) -> Dict[Dict[str, str], Dict[str, np.ndarray]]:
cd = concept_drift_dict()
cd["data"]["is_drift"] = self.expect_return_is_drift
cd["data"]["distance"] = [0.1, 0.2, 0.3]
cd["data"]["p_val"] = [0.1, 0.2, 0.3]
cd["data"]["threshold"] = 0.1
return cd
def predict(self, x: Union[np.ndarray, list], return_p_val: bool = True, return_distance: bool = True) \
-> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
x
Batch of instances.
return_p_val
Whether to return the p-value of the permutation test.
return_distance
Whether to return the LSDD metric between the new batch and reference data.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the p-value, threshold and LSDD metric.
"""
# compute drift scores
p_val, dist, dist_permutations = self.score(x)
drift_pred = int(p_val < self.p_val)
# compute distance threshold
idx_threshold = int(self.p_val * len(dist_permutations))
distance_threshold = np.sort(dist_permutations)[::-1][idx_threshold]
# update reference dataset
if isinstance(self.update_x_ref, dict):
if self.preprocess_fn is not None and self.preprocess_x_ref:
x = self.preprocess_fn(x)
x = self._normalize(x) # type: ignore
elif self.preprocess_fn is None:
x = self._normalize(x) # type: ignore
else:
pass
self.x_ref = update_reference(self.x_ref, x, self.n, self.update_x_ref)
# used for reservoir sampling
self.n += len(x) # type: ignore
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_val
cd['data']['threshold'] = self.p_val
if return_distance:
cd['data']['distance'] = dist
cd['data']['distance_threshold'] = distance_threshold
return cd
def predict(self, x: Union[np.ndarray, list], return_p_val: bool = True,
return_distance: bool = True, return_kernel: bool = True) \
-> Dict[Dict[str, str], Dict[str, Union[int, float, Callable]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
x
Batch of instances.
return_p_val
Whether to return the p-value of the permutation test.
return_distance
Whether to return the MMD metric between the new batch and reference data.
return_kernel
Whether to return the updated kernel trained to discriminate reference and test instances.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the detector's metadata.
'data' contains the drift prediction and optionally the p-value, threshold, MMD metric and
trained kernel.
"""
# compute drift scores
p_val, dist, distance_threshold = self.score(x)
drift_pred = int(p_val < self.p_val)
# update reference dataset
if isinstance(
self.update_x_ref, dict
) and self.preprocess_fn is not None and self.preprocess_x_ref:
x = self.preprocess_fn(x)
self.x_ref = update_reference(
self.x_ref, x, self.n, self.update_x_ref) # type: ignore[arg-type]
# used for reservoir sampling
self.n += len(x)
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_val
cd['data']['threshold'] = self.p_val
if return_distance:
cd['data']['distance'] = dist
cd['data']['distance_threshold'] = distance_threshold
if return_kernel:
cd['data']['kernel'] = self.kernel
return cd
def predict(
self,
x_t: Union[np.ndarray, Any],
return_test_stat: bool = True,
) -> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether the most recent window of data has drifted from the reference data.
Parameters
----------
x_t
A single instance to be added to the test-window.
return_test_stat
Whether to return the test statistic and threshold.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the test-statistic and threshold.
"""
self.t += 1
# preprocess if necessary
if isinstance(self.preprocess_fn, Callable): # type: ignore
x_t = x_t[None, :] if isinstance(x_t, np.ndarray) else [x_t]
x_t = self.preprocess_fn(x_t)[0] # type: ignore
# update test window and return updated test stat
test_stat = self.score(x_t)
threshold = self.get_threshold(self.t)
drift_pred = 0 if test_stat is None else int(test_stat > threshold)
self.test_stats = np.concatenate(
[self.test_stats, np.array([test_stat])])
self.drift_preds = np.concatenate(
[self.drift_preds, np.array([drift_pred])])
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
cd['data']['time'] = self.t
cd['data']['ert'] = self.ert
if return_test_stat:
cd['data']['test_stat'] = test_stat
cd['data']['threshold'] = threshold
return cd
def predict(
self,
x_t: Union[np.ndarray, Any],
return_test_stat: bool = True,
) -> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether the most recent window(s) of data have drifted from the reference data.
Parameters
----------
x_t
A single instance to be added to the test-window(s).
return_test_stat
Whether to return the test statistic and threshold.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the test-statistic and threshold.
"""
# Compute test stat and check for drift
test_stats = self.score(x_t)
thresholds = self.get_threshold(
self.t - 1
) # Note t-1 here, has we wish to use the unconditional thresholds
drift_pred = self._check_drift(test_stats, thresholds)
# Update results attributes
self.test_stats = np.concatenate(
[self.test_stats, test_stats[None, :, :]])
self.drift_preds = np.concatenate(
[self.drift_preds, np.array([drift_pred])])
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
cd['data']['time'] = self.t
cd['data']['ert'] = self.ert
if return_test_stat:
cd['data']['test_stat'] = test_stats
cd['data']['threshold'] = thresholds
return cd
def predict(self, x: np.ndarray, return_p_val: bool = True,
return_distance: bool = True) -> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
x
Batch of instances.
return_p_val
Whether to return the p-value of the test.
return_distance
Whether to return a notion of strength of the drift.
K-S test stat if binarize_preds=False, otherwise relative error reduction.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the performance of the classifier
relative to its expectation under the no-change null.
"""
# compute drift scores
p_val, dist = self.score(x)
drift_pred = int(p_val < self.p_val)
# update reference dataset
if isinstance(self.update_x_ref, dict) and self.preprocess_fn is not None and self.preprocess_x_ref:
x = self.preprocess_fn(x)
self.x_ref = update_reference(self.x_ref, x, self.n, self.update_x_ref)
# used for reservoir sampling
self.n += x.shape[0] # type: ignore
# populate drift dict
# TODO: add instance level feedback
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_val
cd['data']['threshold'] = self.p_val
if return_distance:
cd['data']['distance'] = dist
return cd
def predict(self, X: Union[np.ndarray, list], return_p_val: bool = True,
return_distance: bool = True) -> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
X
Batch of instances.
return_p_val
Whether to return the p-value of the permutation test.
return_distance
Whether to return the MMD metric between the new batch and reference data.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the p-value, threshold and MMD metric.
"""
# compute drift scores
p_val, dist = self.score(X)
drift_pred = int(p_val < self.p_val)
# update reference dataset
if (isinstance(self.update_X_ref, dict) and self.preprocess_fn is not None
and self.preprocess_X_ref):
X = self.preprocess_fn(X)
self.X_ref = update_reference(self.X_ref, X, self.n, self.update_X_ref)
# used for reservoir sampling
self.n += X.shape[0] # type: ignore
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_p_val:
cd['data']['p_val'] = p_val
cd['data']['threshold'] = self.p_val
if return_distance:
cd['data']['distance'] = dist
return cd
def predict(
self,
x_t: Union[np.ndarray, Any],
return_test_stat: bool = True,
) -> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether the most recent window of data has drifted from the reference data.
Parameters
----------
x_t
A single instance to be added to the test-window.
return_test_stat
Whether to return the test statistic and threshold.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the test-statistic and threshold.
"""
# Compute test stat and check for drift
test_stat = self.score(x_t)
threshold = self.get_threshold(self.t)
drift_pred = int(test_stat > threshold)
self.test_stats = np.concatenate(
[self.test_stats, np.array([test_stat])])
self.drift_preds = np.concatenate(
[self.drift_preds, np.array([drift_pred])])
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
cd['data']['time'] = self.t
cd['data']['ert'] = self.ert
if return_test_stat:
cd['data']['test_stat'] = test_stat
cd['data']['threshold'] = threshold
return cd
def predict(self, X: Union[np.ndarray, list], return_metric: bool = True) \
-> Dict[Dict[str, str], Dict[str, Union[int, float]]]:
"""
Predict whether a batch of data has drifted from the reference data.
Parameters
----------
X
Batch of instances.
return_metric
Whether to return the drift metric from the detector.
Returns
-------
Dictionary containing 'meta' and 'data' dictionaries.
'meta' has the model's metadata.
'data' contains the drift prediction and optionally the drift metric and threshold.
"""
# compute drift scores
drift_metric = self.score(X)
drift_pred = int(drift_metric > self.threshold)
# update reference dataset
if isinstance(self.update_X_ref, dict) and self.preprocess_fn is not None and self.preprocess_X_ref:
X = self.preprocess_fn(X)
self.X_ref = update_reference(self.X_ref, X, self.n, self.update_X_ref)
# used for reservoir sampling
self.n += X.shape[0] # type: ignore
# populate drift dict
cd = concept_drift_dict()
cd['meta'] = self.meta
cd['data']['is_drift'] = drift_pred
if return_metric:
cd['data'][self.metric_name] = drift_metric
cd['data']['threshold'] = self.threshold
return cd
