v0.1.4

Added functionality to reset the df back to original copy between executions.
This is to ensure that when the agent is running in a loop it is always working with an intact copy of the dataset discarding any changes like reindexing or column additions made by the previous runs.

v0.1.1

Some further changes to _extract_code() function to sanitize the LLM output

v0.0.8

The _extract_code function has been updated to include a blacklist of Python keywords and functions that are not allowed. These are removed from the extracted code and replaced with comments. These changes help to ensure that potentially dangerous code is not executed by the exec() function. The function also removes any instances of the df = pd.DataFrame() statement with any value inside the parentheses, as this can potentially overwrite the original data.

v0.0.7

Initial release