Street names – r.iresmi.net

Lyon, vue depuis la basilique Notre-Dame de Fourvière — Lyon – CC-BY-NC-ND by Emmanuel Fromm

Day 2 of 30DayMapChallenge: « Lines » (previously). Trying to find the genders in Lyon street names.

We’ll make a map of the street name gender in Lyon. We need a database of french first names where we’ll find the gender. We will extract the Lyon streets from OpenStreetMap.

library(arrow)
library(dplyr)
library(tidyr)
library(readr)
library(purrr)
library(ggplot2)
library(stringr)
library(sf)
library(osmdata)
library(ggspatial)
library(glue)
library(knitr)

set.seed(42)

First names

if (!file.exists("freq_prenoms.rds")) {
  freq_prenoms <- read_parquet("https://www.insee.fr/fr/statistiques/fichier/8205621/prenoms-2023-nat.parquet") |> 
    filter(preusuel != "_PRENOMS_RARES") |> 
    mutate(preusuel = iconv(preusuel, to = "ASCII//TRANSLIT")) |> 
    group_by(preusuel, sexe) |> 
    summarise(n = sum(nombre, na.rm = TRUE),
              .groups = "drop_last") |>
    mutate(total = sum(n)) |> 
    ungroup() |> 
    mutate(sexe = case_when(sexe == 1 ~ "M",
                            sexe == 2 ~ "F",
                            .default = NA_character_)) |> 
    pivot_wider(names_from = sexe, 
                values_from = n,
                values_fill = 0) |> 
    mutate(across(c(M, F), \(x) x / total)) |> 
    write_rds("freq_prenoms.rds")
} else {
  freq_prenoms <- read_rds("freq_prenoms.rds")
}

We have 34234 first names and their gender frequencies since 1900.

Table 1: Sample of first names

preusuel	total	M	F
ZENABOU	48	0	1
EMILIENE	25	0	1
KINGSLEY	878	1	0
DOLOVAN	73	1	0
ERCOLE	67	1	0
YVA	178	0	1
ISSEY	79	1	0
SAWSSEN	121	0	1
MISBAH	24	0	1
GOHANN	20	1	0

Map data

lyon_bbox <- getbb("Lyon, France", featuretype = "city")

if (!file.exists("osm.rds")) {
  lyon <- opq(lyon_bbox) |>
    add_osm_features(features = c(
      '"highway"="motorway"',
      '"highway"="trunk"',
      '"highway"="primary"',
      '"highway"="secondary"',
      '"highway"="tertiary"',
      '"highway"="motorway_link"',
      '"highway"="trunk_link"',
      '"highway"="primary_link"',
      '"highway"="secondary_link"',
      '"highway"="tertiary_link"',
      '"highway"="motorway_junction"',
      '"highway"="unclassified"',
      '"highway"="service"',
      '"highway"="pedestrian"',
      '"highway"="living_street"',
      '"highway"="residential"')) |> 
    osmdata_sf() |> 
    pluck("osm_lines") |> 
    select(osm_id, name) |> 
    drop_na(name) |> 
    group_by(name) |> 
    summarise() |> 
    write_rds("osm.rds")
} else {
  lyon <- read_rds("osm.rds")
}

That’s 3270 street names.

Finding first names in street names

We use a brute-force method: for each street we check if a part of its label is present in our list of female or male first names. We keep only first names with a high frequency in any of the genders.

female <- freq_prenoms |> 
  filter(F > .8,
         str_length(preusuel) > 1,
         preusuel != "LA") |> 
  pull(preusuel)

male <- freq_prenoms |> 
  filter(M > .8, 
         str_length(preusuel) > 1) |> 
  pull(preusuel)

street_gender <- lyon |> 
  mutate(name = str_to_upper(iconv(name, to = "ASCII//TRANSLIT")),
         f = str_extract_all(name, glue_collapse(female, sep = "\\b|\\b", last = "\\b")),
         m = str_extract_all(name, glue_collapse(male, sep = "\\b|\\b", last = "\\b")),
         gender = unlist(
           map2(f, m, 
                ~ case_when(length(.x) > length(.y)       ~ "female",
                            length(.y) > length(.x)       ~ "male",
                            identical(.x, character(0)) & 
                              identical(.y, character(0)) ~ "not concerned",
                            length(.x) == length(.y)      ~ "undecidable",
                            .default = NA_character_))))

Table 2: Classification sample

name	geometry	f	m	gender
COURS DE VERDUN RECAMIER	LINESTRING (4.830426 45.748…			not concerned
IMPASSE DES ANGLAIS	LINESTRING (4.795807 45.753…			not concerned
RUE DES PROVENCES	LINESTRING (4.79335 45.7369…			not concerned
CHEMIN DES PEUPLIERS	LINESTRING (4.866587 45.801…			not concerned
ALLEE DU LEVANT	LINESTRING (4.878859 45.759…			not concerned
RUE ROPOSTE	LINESTRING (4.866353 45.760…			not concerned
ALLEE NELLIE BLY	LINESTRING (4.84882 45.7429…	NELLIE		female
QUAI JEAN MOULIN	MULTILINESTRING ((4.837853 …		JEAN	male
LA VIEILLE ROUTE	LINESTRING (4.769782 45.720…			not concerned
AVENUE DE CHAMPAGNE	MULTILINESTRING ((4.796801 …			not concerned

Map

pal_mf <- c("female" = "lightpink1",
            "male" = "lightskyblue",
            "undecidable" = "lightyellow4",
            "not concerned" = "seashell2")

street_gender |> 
  mutate(gender = factor(gender, levels = names(pal_mf))) |> 
  st_set_crs("EPSG:4326") |> 
  ggplot() +
  geom_sf(aes(color = gender), 
          linewidth = .5,
          key_glyph = "timeseries") +
  scale_color_manual(values = pal_mf) +
  annotation_scale(bar_cols =  c("darkgrey", "white"),
                   line_col = "darkgrey",
                   text_col = "darkgrey",
                   height = unit(0.1, "cm")) +
  coord_sf(xlim = lyon_bbox[c(1, 3)],
           ylim = lyon_bbox[c(2, 4)]) +
  labs(title = "Gender in Lyon street names",
       color = "",
       caption = glue("Map data © OpenStreetMap contributors
                      using INSEE Fichier des prénoms 2023
                      r.iresmi.net - {Sys.Date()}")) +
  theme_void() +
  theme(plot.background = element_rect(color = NA, 
                                       fill = "white"),
        plot.caption = element_text(size = 5,
                                    color = "darkgrey"))

A map of gender in Lyon street names — Figure 1: Lyon

Possible miss-classifications

Lots of bias make this map unreliable, and would need manual editing…

epicenous first names

some first names can be male or female (GWEN, CAMILLE, DOMINIQUE)

not concerned

street names of people but without the first name (RUE VILLON),
title instead of first name (RUE DE L’AMIRAL COURBET),

has a gender but shouldn’t

common names used as first name (CHEMIN DE LA POMME), mainly for girls…
strange first names (AUTOROUTE DU SOLEIL, Soleil seems to be a girl name…)

accidentally well classified

the last name is also a first name (COURS BAYARD)

--- title: "Street names" description: "Day 2 of 30DayMapChallenge. Lines" author: "Michaël" date: "2024-11-02T14:00:00" date-modified: last-modified categories: - R - spatial - datavisualization - Parquet - OSM - 30DayMapChallenge draft: false freeze: true editor_options: chunk_output_type: console chunk_output_type: console execute: eval: true --- ```{r} #| label: preview #| echo: false #| fig-cap: "Lyon -- CC-BY-NC-ND by Emmanuel Fromm" #| fig-alt: "Lyon, vue depuis la basilique Notre-Dame de Fourvière" #| fig-link: https://www.flickr.com/photos/isobrown/5506682247 #| class-output: "preview-image" #| out-width: "100%" knitr::include_graphics("images/5506682247_bf17b1ae0b_c.jpg") ``` Day 2 of [30DayMapChallenge](https://30daymapchallenge.com/): « Lines » ([previously](/#category=30DayMapChallenge)). Trying to find the genders in Lyon street names. We'll make a map of the street name gender in [Lyon](https://en.wikipedia.org/wiki/Lyon). We need a database of [french first names](https://www.insee.fr/fr/statistiques/8205621) where we'll find the gender. We will extract the Lyon streets from [OpenStreetMap](https://www.openstreetmap.org/#map=12/45.7533/4.8927). ```{r} #| label: config library(arrow) library(dplyr) library(tidyr) library(readr) library(purrr) library(ggplot2) library(stringr) library(sf) library(osmdata) library(ggspatial) library(glue) library(knitr) set.seed(42) ``` ## First names ```{r} #| label: firstnames-data if (!file.exists("freq_prenoms.rds")) { freq_prenoms <- read_parquet("https://www.insee.fr/fr/statistiques/fichier/8205621/prenoms-2023-nat.parquet") |> filter(preusuel != "_PRENOMS_RARES") |> mutate(preusuel = iconv(preusuel, to = "ASCII//TRANSLIT")) |> group_by(preusuel, sexe) |> summarise(n = sum(nombre, na.rm = TRUE), .groups = "drop_last") |> mutate(total = sum(n)) |> ungroup() |> mutate(sexe = case_when(sexe == 1 ~ "M", sexe == 2 ~ "F", .default = NA_character_)) |> pivot_wider(names_from = sexe, values_from = n, values_fill = 0) |> mutate(across(c(M, F), \(x) x / total)) |> write_rds("freq_prenoms.rds") } else { freq_prenoms <- read_rds("freq_prenoms.rds") } ``` We have `r nrow(freq_prenoms)` first names and their gender frequencies since 1900. ```{r} #| label: tbl-firstnames-sample #| echo: false #| tbl-cap: "Sample of first names" freq_prenoms |> slice_sample(n = 10) |> kable() ``` ## Map data ```{r} #| label: osm-data lyon_bbox <- getbb("Lyon, France", featuretype = "city") if (!file.exists("osm.rds")) { lyon <- opq(lyon_bbox) |> add_osm_features(features = c( '"highway"="motorway"', '"highway"="trunk"', '"highway"="primary"', '"highway"="secondary"', '"highway"="tertiary"', '"highway"="motorway_link"', '"highway"="trunk_link"', '"highway"="primary_link"', '"highway"="secondary_link"', '"highway"="tertiary_link"', '"highway"="motorway_junction"', '"highway"="unclassified"', '"highway"="service"', '"highway"="pedestrian"', '"highway"="living_street"', '"highway"="residential"')) |> osmdata_sf() |> pluck("osm_lines") |> select(osm_id, name) |> drop_na(name) |> group_by(name) |> summarise() |> write_rds("osm.rds") } else { lyon <- read_rds("osm.rds") } ``` That's `{r} nrow(lyon)` street names. ## Finding first names in street names We use a brute-force method: for each street we check if a part of its label is present in our list of female or male first names. We keep only first names with a high frequency in any of the genders. ```{r} #| label: classification female <- freq_prenoms |> filter(F > .8, str_length(preusuel) > 1, preusuel != "LA") |> pull(preusuel) male <- freq_prenoms |> filter(M > .8, str_length(preusuel) > 1) |> pull(preusuel) street_gender <- lyon |> mutate(name = str_to_upper(iconv(name, to = "ASCII//TRANSLIT")), f = str_extract_all(name, glue_collapse(female, sep = "\\b|\\b", last = "\\b")), m = str_extract_all(name, glue_collapse(male, sep = "\\b|\\b", last = "\\b")), gender = unlist( map2(f, m, ~ case_when(length(.x) > length(.y) ~ "female", length(.y) > length(.x) ~ "male", identical(.x, character(0)) & identical(.y, character(0)) ~ "not concerned", length(.x) == length(.y) ~ "undecidable", .default = NA_character_)))) ``` ```{r} #| label: tbl-classification-sample #| echo: false #| tbl-cap: "Classification sample" street_gender |> slice_sample(n = 10) |> kable() ``` ## Map ```{r} #| label: fig-map #| fig-cap: "Lyon" #| fig-alt: "A map of gender in Lyon street names" #| out-width: "100%" pal_mf <- c("female" = "lightpink1", "male" = "lightskyblue", "undecidable" = "lightyellow4", "not concerned" = "seashell2") street_gender |> mutate(gender = factor(gender, levels = names(pal_mf))) |> st_set_crs("EPSG:4326") |> ggplot() + geom_sf(aes(color = gender), linewidth = .5, key_glyph = "timeseries") + scale_color_manual(values = pal_mf) + annotation_scale(bar_cols = c("darkgrey", "white"), line_col = "darkgrey", text_col = "darkgrey", height = unit(0.1, "cm")) + coord_sf(xlim = lyon_bbox[c(1, 3)], ylim = lyon_bbox[c(2, 4)]) + labs(title = "Gender in Lyon street names", color = "", caption = glue("Map data © OpenStreetMap contributors using INSEE Fichier des prénoms 2023 r.iresmi.net - {Sys.Date()}")) + theme_void() + theme(plot.background = element_rect(color = NA, fill = "white"), plot.caption = element_text(size = 5, color = "darkgrey")) ``` ```{r} #| label: fig-map-leaflet #| fig-cap: "Lyon" #| fig-alt: "An interactive map of gender in Lyon street names" #| out-width: "100%" #| eval: false #| echo: false library(leaflet) pal <- colorFactor(pal_mf, levels = names(pal_mf)) street_gender |> leaflet() |> addTiles() |> addPolylines(popup = ~name, opacity = 1, weight = 3, color = ~ pal(gender)) |> addLegend(pal = pal, values = names(pal_mf)) ``` ## Possible miss-classifications Lots of bias make this map unreliable, and would need manual editing... ### epicenous first names - some first names can be male or female (GWEN, CAMILLE, DOMINIQUE) ### not concerned - street names of people but without the first name (RUE VILLON), - title instead of first name (RUE DE L'AMIRAL COURBET), ### has a gender but shouldn't - common names used as first name (CHEMIN DE LA POMME), mainly for girls... - strange first names (AUTOROUTE DU SOLEIL, Soleil seems to be a girl name...) ### accidentally well classified - the last name is also a first name (COURS BAYARD)