CCA - 003: Rolling window segmentation

[saumyadholia25-source]

Implemented rolling window logic and correlation computation for multi-sensor time-series data

[senuradp]

Good work on exploring rolling window segmentation and validating it across different datasets. The approach and logic are correct and relevant for the correlation alert pipeline.

However, for integration into the project, we need this to follow the modular structure defined in the wrapper.

Please refactor your implementation into a single function:
create_rolling_windows(df, window_size, step_size)

This function should:

• take the preprocessed dataframe as input
• return a list of windowed dataframes
• avoid implementing other pipeline stages (correlation, alerts, etc.)
• avoid notebook-based execution in the final version

The notebooks are useful for experimentation, but the final PR should contain a clean Python implementation that fits into the wrapper pipeline.

[saumyadholia25-source]

rolling_windows.py.ipynb

uploaded the updated file

[senuradp]

Thanks for updating the PR based on the earlier feedback. I reviewed the updated rolling window implementation, and this version is much better aligned with the intended modular pipeline structure.

The create_rolling_windows(df, window_size, step_size) function now focuses only on rolling window segmentation, accepts the expected inputs, and returns a list of windowed DataFrames. The code is clean, readable, and no longer overlaps with other pipeline stages.

There is one minor issue to fix before final integration: the function currently resets the index inside each window using reset_index(drop=True). Since the preprocessed dataframe uses the timestamp as the index, we should preserve the original index so the correlation stage can correctly capture the start_time and end_time for each window.

I updated:
window = df.iloc[start:end].copy().reset_index(drop=True)

to:
window = df.iloc[start:end].copy()

Overall, this is a solid rolling window implementation and is ready to proceed after this minor fix.