Improve race name normalisation and/or partial matching

[k-j-m]

Race results should be able to be matched to the corresponding race information despite slight differences in the race name that is used in either dataset so that all results can be matched against races. When building an index of race results to race information, I've found that ~50% of race results can't be matched.

• Error raised when initially scraping the race info means that the race info isn't available
• Inconsistent apostrophe's in race names
• Inconsistent capitalisation in race names
• Inconsistent use of ordinals in race names (eg 62nd Three Peaks)
• Inconsistently using the 'Race' suffix, eg 'Kenny Malton Memorial' vs 'Kenny Malton Memorial Race'
• Age group results available, but race info not available
• Inconsistent use of 'Hill' suffix, eg 'Manor Water' vs 'Manor Water Hill'
• Champs races causing inconsistent results, eg 'Shutlingsloe & Junior champs races' vs 'Shutlingsloe'
• 'Isle of' prefix used inconsistently, eg 'Isle of Jura' vs 'Jura'
• Memorial prefix used inconsistently, eg 'Don Morrison Memorial Edale Skyline' vs 'Edale Skyline'

[k-j-m]

We can attack this in a number of ways:

• log all race info pages which raise an error of some sort which ISN'T specifically due to an 'Unknown' race category
• remove all apostrophes from names
• convert all names to lower case
• where 'race' is the last word, drop it
• where 'memorial' is the last word, drop it
• where 'memorial' is in the middle of the name, take only what comes afterwards
• where 'hill' is the last word, drop it
• where the first two words are 'isle of', drop them
• where the first word is an ordinal (eg 62nd), drop it Apart from the first point, all of these can be done locally without touching the website - so should be v.quick work. Try incrementally implementing each of these schemes and note how the 'match rate' improves with each.

[k-j-m]

For the 'Shutlingsloe & junior champs' example something smarter would be needed. If there are sufficiently few of these leftovers then we can just do a manual job of adding the race id + result id to the index file.

[k-j-m]

Crumbs, I've only managed to improve from 58% to 60% hit rate. Time to look at for some more specific examples to see what the most common problems are.

[k-j-m]

Got it up to 65% - most of the remaining mismatches seem to be either because the race info doesn't exist, or because we have results for different classification (not worth trying to use).