anomalies: Empty DataFrame

[Aquariuaa]

Hello! Thank you very much for your contribution to LLM4TS. When I run the code you provided, I find that the returned results are always null on both the example data you provided and on multiple TS anomaly detection datasets (SWAP)，the returned as follow:

anomalies: Empty DataFrame
Columns: [start, end, severity]
Index: []

This means that SigLLM is almost impossible to detect anomalies. My hyperparameters are set as follows, is there anything in my code that doesn't make sense for the hyperparameters?

pipeline_name = 'gpt_detector'
hyperparameters = {
"sigllm.primitives.forecasting.custom.rolling_window_sequences#1": {
"target_column": 0,
"window_size": 140,
"step_size": 1,
"target_size": 1
},

"orion.primitives.timeseries_anomalies.find_anomalies#1": {
    "fixed_threshold": True,
    "anomaly_padding": 2
},}

[sarahmish]

Hi @Aquariuaa – thank you for using sigllm!

The example we show in the README should work in most cases. I created a colab notebook to help test the example out fairy quickly.

There are two critical attributes to set based on your data:

• interval: which describes the number of timestamps between one entry and another.
• decimal: which sets the number of decimal points to keep within the data when transforming it to an LLM-ready value.

Hope this helps explore the right configuration for your data!