Unexpected inplace dtype conversion

[FabianHofmann]

The function dedupe_dataframe changes the dtypes of the input dataframe (e.g. float to string). One would rather expect this functions to not change the input arguments. One could prevent this by modifying a copy of the original dataframe.

[Lyonk71]

Thanks for the comment. Implementing your suggestion would provide more abstraction than is my intention. In fact, I actually dislike that the program currently abstracts the pre-processing step (which converts float to string). In the future, I may separate out the pre-processing step to make that conversion process explicit. What do you think?

IIWY, I'd generate a hash ID for each record prior to running pandas-dedupe, then join accordingly afterwards - it should only be a few lines of code. Let me know if you need more guidance on that.

[FabianHofmann]

Thanks for the answer @Lyonk71, I rather thought of a copy of all columns needed for clustering (and keep the pre-processing steps hidden), like

    cols = [c if isinstance(c, str) else c[0] for c in field_properties]
    data = df[cols]

inserted here

pandas-dedupe/pandas_dedupe/dedupe_dataframe.py

Line 238 in a4abc25

and then use data instead of df in the subsequent code. Any downside I oversaw? If you want I could make a PR