Allow detectors to work on a dataframe with 1 target and multiple passive columns about adtk HOT 1 OPEN

@szeitlin Sorry I don't think I get your question quite well. But here are a few things that I guess may be relevant to what you asked.

1. All adtk detectors return binary pandas Series with the DateTime index as the input Series unless option return_list is turned True (by default it is False).

2. For univariate detectors (ThresholdAD, QuantileAD, InterQuartileRangeAD, GeneralizedESDTestAD, PersistAD, LevelShiftAD, VolatilityShiftAD, AutoregressionAD, SeasonalAD, CustomizedDetector1D), if a pandas DataFrame is given as input, the detector will be applied to every column independently and the output will be a DataFrame too (unless option return_list is set as True). A similar design applies to transformers too. If a user has a df with column "A", "B", "C" but only wants to apply a model to "A" and "C", we assume s/he will do model.detect(df[["A", "C"]]).

3. For multivariate detectors (MinClusterDetector, OutlierDetector, RegressionAD, PcaAD, CustomizedDetectorHD), the input must be pandas DataFrame and a Series will trigger a TypeError. A similar design applies to transformers too.

4. If a user needs to apply different adtk models to different columns of a single DataFrame object and then aggregate the anomalies, it is recommended to use Pipenet and the option of "subset" in parameter steps could be helpful.

I don't know if this helps. If not, we'd appreciate it if you may offer an example to describe your question. Thanks 😃

from adtk.