score anomalies

path: orion.primitives.tadgan.score_anomalies

description: this primitive computes an array of anomaly scores based on a combination of reconstruction error and critic output.

see json.

argument	type	description
parameters
y	numpy.ndarray	ground truth
y_hat	numpy.ndarray	predicted values. Each timestamp has multiple predictions
critic	numpy.ndarray	critic score. Each timestamp has multiple critic scores
index	numpy.ndarray	time index for each y (start position of the window)
hyperparameters
score_window	int	size of the window over which the scores are calculated. If not given, 10 is used
critic_smooth_window	int	size of window over which smoothing is applied to critic. If not given, 200 is used
error_smooth_window	int	size of window over which smoothing is applied to error. If not given, 200 is used.
rec_error_type	str	the method to compute reconstruction error. Can be one of [“point”, “area”, “dtw”]
comb	str	how to combine critic and reconstruction error. Can be one of [“mult”, “sum”, “rec”]
lambda_rec	float	used if comb=”sum” as a lambda weighted sum to combine scores. If not given, 0.5 is used.
output
errors	numpy.ndarray	array of anomaly scores
true_index	numpy.ndarray	time index of errors
true	numpy.ndarray	ground truth
predictions	numpy.ndarray	predicted sequence

In [1]: import numpy as np

In [2]: from mlstars import load_primitive

In [3]: primitive = load_primitive('orion.primitives.tadgan.score_anomalies',
   ...:     arguments={"error_smooth_window": 10, "critic_smooth_window": 10,
   ...:                "score_window": 10, "comb": "rec"})
   ...: 

In [4]: y = np.array([1] * 100).reshape(1, -1, 1)

In [5]: y_hat = [0.9] * 40 + [0.5] * 10 + [1.1] * 10 + [0.99] * 40

In [6]: y_hat = np.array(y_hat).reshape(1, -1, 1)

In [7]: critic = np.array([[0.5]])

In [8]: index = np.array([[1]])

In [9]: errors, true_index, true, predictions = primitive.produce(
   ...:     y=y, y_hat=y_hat, critic=critic, index=index)
   ...: 

In [10]: print("average error value: {:.2f}".format(errors.mean()))
average error value: 1.33

VAE reconstuction errors