What is going on in undocumented first demo (https://github.com/datamllab/tods#examples-and-tutorials and demo: General Usage)

[paapu88]

Dear Developers,
I try to make sense of the first demo available in your www page (General Usage)

There one reads in dataset 500_UCR_Anomaly_robotDOG1_10000_19280_19360.txt
Which contains 20000 floats or 40000 floats, depending where one takes the file.
There are no labels.

In the notebook one predicts based on 10000 first values with some blackbox method (DeepLogSKI)
and for the last 10000 values:

X_train = np.expand_dims(data_UCR[:10000], axis=1)
X_test = np.expand_dims(data_UCR[10000:], axis=1)
transformer = DeepLogSKI()
transformer.fit(X_train)
prediction_labels_train = transformer.predict(X_train)
prediction_labels_test = transformer.predict(X_test)
prediction_score = transformer.predict_score(X_test)
y_true = prediction_labels_train
y_pred = prediction_labels_test
precision, recall, thresholds = precision_recall_curve(y_true, y_pred)

Some questions:

1. how can one get true values from predictions?

prediction_labels_train = transformer.predict(X_train)

2. Even if those values would somehow be "true" how can one compare those to totally different part of the dataset?

prediction_labels_test = transformer.predict(X_test)

3. Any documentation for DeepLogSKI() ?
4. what is prediction_score:

prediction_score = transformer.predict_score(X_test)

Terveisin, Markus

[paapu88]

Closed due to inactivity. This seems a pretty dead repo, requires python <=3.8 etc, is based on darpa blackboxes.