Postcode comparison template

[zslade]

Addresses issue #215

Comparison template for postcode column. The default arguments will give a comparison with levels:
- Exact match on full postcode
- Exact match on sector
- Exact match on district
- Exact match on area
- All other comparisons

with an optional 'distance in km' comparison level

[github-actions]

Test: test_2_rounds_1k_duckdb

Percentage change: -28.0%

	date	time	stats_mean	stats_min	commit_info_branch	commit_info_id	machine_info_cpu_brand_raw	machine_info_cpu_hz_actual_friendly	commit_hash
849	2022-07-12	18:40:05	1.89098	1.87463	splink3	c334bb9	Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz	2.7934 GHz	c334bb9
1635	2023-05-18	15:27:31	1.37419	1.3496	(detached head)	9039d3a	Intel(R) Xeon(R) Platinum 8272CL CPU @ 2.60GHz	2.5939 GHz	9039d3a

Test: test_2_rounds_1k_sqlite

Percentage change: -24.0%

	date	time	stats_mean	stats_min	commit_info_branch	commit_info_id	machine_info_cpu_brand_raw	machine_info_cpu_hz_actual_friendly	commit_hash
851	2022-07-12	18:40:05	4.32179	4.25898	splink3	c334bb9	Intel(R) Xeon(R) Platinum 8370C CPU @ 2.80GHz	2.7934 GHz	c334bb9
1637	2023-05-18	15:27:31	3.23967	3.23781	(detached head)	9039d3a	Intel(R) Xeon(R) Platinum 8272CL CPU @ 2.60GHz	2.5939 GHz	9039d3a

Click here for vega lite time series charts

[RossKen]

@zslade this looks good at first look (but currently on my phone so will look at properly tomorrow).

One thing I think should be added is the ability to add multiple km distance levels.

You can see an example of how to deal with multiple levels at

splink/splink/comparison_template_library.py

Line 195 in 6e3a2e1

levenshtein_thresholds = ensure_is_iterable(levenshtein_thresholds)

So you will need to use ensure_is_iterable then distance_threshold_comparison_levels to get this working for the levels themselves.

Plus add to the comparison description similar to

splink/splink/comparison_template_library.py

Line 275 in 6e3a2e1

if len(levenshtein_thresholds) > 0:

With distance_threshold_description

[RossKen]

Also, it would be good to update the feature engineering topic guide with this function for the postcode section

[RossKen]

And there are postcode columns in some of the splink demos which should be updated with this function instead

[RossKen]

Hey @zslade , I was just having a skim through the fe docs. Apologies, I should have been clearer on what I was thinking here. The focus of that topic guide is on what adding additional columns can do to improve a splink model, so the one around postcodes was mainly looking at how adding lat/long could add additional levels to match on.

So here, all I was thinking was to replace the initial levenshtein_at_thresholds with postcode_comparison then changing some of the narrative around it to talk about the benefits of adding distance_in_km levels ontop the purely postcode solution. I.e. being able to compare places that are close, but in different postcode regions e.g. N London postcodes vs SW London postcodes are right beside each other but the postcode wouldn't tell you that on their own.

Apologies for any confusion - that's my bad!

[RossKen]

Great work! I just pushed some small changes, and now happy for you to merge! 🎉