def __init__(self,
x_ref: Union[np.ndarray, list],
ert: float,
window_size: int,
backend: str = 'tensorflow',
preprocess_fn: Optional[Callable] = None,
sigma: Optional[np.ndarray] = None,
n_bootstraps: int = 1000,
n_kernel_centers: Optional[int] = None,
lambda_rd_max: float = 0.2,
device: Optional[str] = None,
verbose: bool = True,
input_shape: Optional[tuple] = None,
data_type: Optional[str] = None) -> None:
"""
Online least squares density difference (LSDD) data drift detector using preconfigured thresholds.
Motivated by Bu et al. (2017): https://ieeexplore.ieee.org/abstract/document/7890493
We have made modifications such that a desired ERT can be accurately targeted however.
Parameters
----------
x_ref
Data used as reference distribution.
ert
The expected run-time (ERT) in the absence of drift. For the multivariate detectors, the ERT is defined
as the expected run-time from t=0.
window_size
The size of the sliding test-window used to compute the test-statistic.
Smaller windows focus on responding quickly to severe drift, larger windows focus on
ability to detect slight drift.
backend
Backend used for the LSDD implementation and configuration.
preprocess_fn
Function to preprocess the data before computing the data drift metrics.s
sigma
Optionally set the bandwidth of the Gaussian kernel used in estimating the LSDD. Can also pass multiple
bandwidth values as an array. The kernel evaluation is then averaged over those bandwidths. If `sigma`
is not specified, the 'median heuristic' is adopted whereby `sigma` is set as the median pairwise distance
between reference samples.
n_bootstraps
The number of bootstrap simulations used to configure the thresholds. The larger this is the
more accurately the desired ERT will be targeted. Should ideally be at least an order of magnitude
larger than the ert.
n_kernel_centers
The number of reference samples to use as centers in the Gaussian kernel model used to estimate LSDD.
Defaults to 2*window_size.
lambda_rd_max
The maximum relative difference between two estimates of LSDD that the regularization parameter
lambda is allowed to cause. Defaults to 0.2 as in the paper.
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.
verbose
Whether or not to print progress during configuration.
input_shape
Shape of input data.
data_type
Optionally specify the data type (tabular, image or time-series). Added to metadata.
"""
super().__init__()
backend = backend.lower()
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'MMDDrift detector with {backend} backend.')
elif backend not in ['tensorflow', 'pytorch']:
raise NotImplementedError(
f'{backend} not implemented. Use tensorflow or pytorch instead.'
)
kwargs = locals()
args = [kwargs['x_ref'], kwargs['ert'], kwargs['window_size']]
pop_kwargs = [
'self', 'x_ref', 'ert', 'window_size', 'backend', '__class__'
]
[kwargs.pop(k, None) for k in pop_kwargs]
if backend == 'tensorflow' and has_tensorflow:
kwargs.pop('device', None)
self._detector = LSDDDriftOnlineTF(*args, **kwargs) # type: ignore
else:
self._detector = LSDDDriftOnlineTorch(*args,
**kwargs) # type: ignore
self.meta = self._detector.meta
def test_lsdd_online(lsdd_online_params):
n_features, ert, window_size, preprocess, n_bootstraps = lsdd_online_params
np.random.seed(0)
torch.manual_seed(0)
x_ref = np.random.randn(n * n_features).reshape(n, n_features).astype(
np.float32)
preprocess_fn, preprocess_kwargs = preprocess
to_list = False
if hasattr(preprocess_fn,
'__name__') and preprocess_fn.__name__ == 'preprocess_list':
to_list = True
x_ref = [_[None, :] for _ in x_ref]
elif isinstance(preprocess_fn, Callable) and 'layer' in list(preprocess_kwargs.keys()) \
and preprocess_kwargs['model'].__name__ == 'HiddenOutput':
model = MyModel(n_features)
layer = preprocess_kwargs['layer']
preprocess_fn = partial(preprocess_fn,
model=HiddenOutput(model=model, layer=layer))
else:
preprocess_fn = None
cd = LSDDDriftOnlineTorch(x_ref=x_ref,
ert=ert,
window_size=window_size,
preprocess_fn=preprocess_fn,
n_bootstraps=n_bootstraps)
x_h0 = np.random.randn(n * n_features).reshape(n, n_features).astype(
np.float32)
detection_times_h0 = []
test_stats_h0 = []
for x_t in x_h0:
if to_list:
x_t = [x_t]
pred_t = cd.predict(x_t, return_test_stat=True)
test_stats_h0.append(pred_t['data']['test_stat'])
if pred_t['data']['is_drift']:
detection_times_h0.append(pred_t['data']['time'])
cd.reset()
average_delay_h0 = np.array(detection_times_h0).mean()
test_stats_h0 = [ts for ts in test_stats_h0 if ts is not None]
assert ert / 3 < average_delay_h0 < 3 * ert
cd.reset()
x_h1 = 1 + np.random.randn(n * n_features).reshape(n, n_features).astype(
np.float32)
detection_times_h1 = []
test_stats_h1 = []
for x_t in x_h1:
if to_list:
x_t = [x_t]
pred_t = cd.predict(x_t, return_test_stat=True)
test_stats_h1.append(pred_t['data']['test_stat'])
if pred_t['data']['is_drift']:
detection_times_h1.append(pred_t['data']['time'])
cd.reset()
average_delay_h1 = np.array(detection_times_h1).mean()
test_stats_h1 = [ts for ts in test_stats_h1 if ts is not None]
assert np.abs(average_delay_h1) < ert / 2
assert np.mean(test_stats_h1) > np.mean(test_stats_h0)