Not able to run Timeseries Explainer incase of anomaly detection

[usnishmukherjee]

I am trying to perform Explainability for Timeseries Anomaly detection ( DBSCAN model). I have loaded the model and prediction funtion for DBSCAN is model.fit_predict
Here is the code snippet:

And the Final errror : ValueError: Unable to convert array of bytes/strings into decimal numbers with dtype='numeric'
Although my dataframe contains only numeric values.
Here is the view of my data:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 300 entries, 0 to 299
Data columns (total 1 columns):
Column Non-Null Count Dtype
sample 300 non-null float64
dtypes: float64(1)

[usnishmukherjee]

I am able to run the shapExplainer for a timeseries data. But not sure how to read the graph. Not enough information is given in the documentation as well.
Please help to understand the graph.

What to infer from the graph? What is the anomaly and why that is anomaly? I have passed a batch of data here Like below:
explanations = explainer.explain(Timeseries.from_pd(test_df))

[yangwenz]

Hi, the score (value_score) shows the feature importance score of each timestamp if SHAP is applied. Similarly, the timestamp_score is the feature importance score of the timestamp. The timestamp_score will be zero if the timestamps are not used during anomaly detection.

[usnishmukherjee]

Hi @yangwenz .
Thanks for your help. One more thing, how that importance score help me to identify the anomaly ?

[yangwenz]

Unfortunately, the lib cannot identify the anomaly, it can only explain why a certain data point or a window is detected as anomaly. For example, when you apply some black-box anomaly detector (especially those based on neural networks), you may need to know why a window of timeseries is detected as an anamaly and use this information to analyze the corresponding incident. This is quite common in AIOps.