regression in result quality

[joostschouppe]

In recent geocoding on data that I geocoded before, I see a regression in quality. It seems to be related to the language of the street name. If you are in Brussels and give a Dutch streetname, or in the German speaking area and give a German street name, the results seem to be a lot worse then when you give a French streetname. This was not an issue a year or so ago. In one dataset, I have Brussels addresses with both the Dutch and the French streetname. They have a 8% success rate in Dutch and 96% success in French.

[hugoperilleux]

Thank you for your feedback. Sorry for the inconvenience! We've restored phacochr to its previous version (0.9.14). If you do:

devtools::install_github(‘phacochr/phacochr’)
library(phacochr)
phaco_setup_data()

and optionally:

phaco_best_data_update()

it should work as before.

Also, would you be willing to share your address set with me so that I can try it out (hugo.perilleux@ulb.be)?

[joostschouppe]

Yes, this solves the issue. Here are some test results:

Some Dutch in Brussels, some German in Ostbelgien	Region	n	Valid rue(%)	Rue detect.(%valid)	Approx.(n)	Elarg.(n)	Mid.(n)	Geocode(%valid)	Geocode(%tot)
Bruxelles	26	100	100.0	1	0	0	100.0	100.0
Flandre	70	100	97.1	3	0	2	97.1	97.1
Wallonie	54	100	90.7	6	1	0	88.9	88.9
NA	2	100	0.0	0	0	0	0.0	0.0
Total	152	100	94.1	10	1	2	93.4	93.4

(here the results were about 50% in Brussels before)

French where available	Region	n	Valid rue(%)	Rue detect.(%valid)	Approx.(n)	Elarg.(n)	Mid.(n)	Geocode(%valid)	Geocode(%tot)
Bruxelles	26	100	96.2	1	0	0	96.2	96.2
Flandre	70	100	97.1	3	0	2	97.1	97.1
Wallonie	54	100	90.7	6	1	0	88.9	88.9
NA	2	100	0.0	0	0	0	0.0	0.0
Total	152	100	93.4	10	1	2	92.8	92.8