Skip to content

Commit

Permalink
sample addresses per postcode weighted by stroke incidence estimates
Browse files Browse the repository at this point in the history
  • Loading branch information
@mpadge
mpadge committed Feb 12, 2019
1 parent ea969ee commit 35c98ca
Show file tree
Hide file tree
Showing 2 changed files with 72 additions and 54 deletions.
37 changes: 23 additions & 14 deletions RehabCatchment/rvspy/README.Rmd
View file
Original file line number Diff line number Diff line change
Expand Up @@ -95,15 +95,6 @@ basicDemographicsRehab <- filter(basicDemographicsVIC,
mutate(Postcode = as.numeric(POA_CODE16)) %>%
select(-starts_with("POA_"))
```
The python code has fewer postcodes than the R code, with numbers determined
manually here by comparing the corresponding maps. The reduced version
equivalent to the python code is:
```{r}
#mapview::mapview (basicDemographicsRehab_py)
removes <- c (40, 56, 57, 53, 43, 10, 7, 8, 29, 11, 1, 3)
index <- seq (nrow (basicDemographicsRehab))
basicDemographicsRehab_py <- basicDemographicsRehab [!index %in% removes, ]
```

## Data sampling

Expand All @@ -114,8 +105,26 @@ do this, and so sample random points from within the postcode polygons. These
two approaches are replicated here in R code, the first referred to as
`randomaddresses`, the second as `randomPoints`.

The `addressesPerPostcode` value below is modified by the estimated stroke rate
per postcode calculated in the python code.

```{r addressesPerPostcode}
addressesPerPostcode <- 1000
addressesPerPostcode <- 10
stroke_rate <- read.csv ("../../python/notebooks/data/postcode_strokes.csv",
stringsAsFactors = FALSE)
stroke_rate$POSTCODE <- substr (stroke_rate$POA_CODE, 4, 7)
stroke_rate$strokes <- round (stroke_rate$strokes * addressesPerPostcode)
index <- match (basicDemographicsRehab$Postcode, stroke_rate$POSTCODE)
basicDemographicsRehab$stroke_rate <- stroke_rate$strokes [index]
```
The python code has fewer postcodes than the R code, with numbers determined
manually here by comparing the corresponding maps. The reduced version
equivalent to the python code is:
```{r}
#mapview::mapview (basicDemographicsRehab_py)
removes <- c (40, 56, 57, 53, 43, 10, 7, 8, 29, 11, 1, 3)
index <- seq (nrow (basicDemographicsRehab))
basicDemographicsRehab_py <- basicDemographicsRehab [!index %in% removes, ]
```

Random addresses:
Expand All @@ -128,14 +137,14 @@ samplePCode <- function(pcode, number) {
randomaddresses <- map(basicDemographicsRehab$Postcode,
samplePCode,
number=addressesPerPostcode) %>%
number=basicDemographicsRehab$stroke_rate) %>%
bind_rows() %>%
sf::st_as_sf(coords = c("LONGITUDE", "LATITUDE"),
crs=st_crs(basicDemographicsRehab),
agr = "constant")
randomaddresses_py <- map(basicDemographicsRehab_py$Postcode,
samplePCode,
number=addressesPerPostcode) %>%
number=basicDemographicsRehab$stroke_rate) %>%
bind_rows() %>%
sf::st_as_sf(coords = c("LONGITUDE", "LATITUDE"),
crs=st_crs(basicDemographicsRehab),
Expand All @@ -146,15 +155,15 @@ Random points:
```{r random-points}
randomPoints <- apply (basicDemographicsRehab, 1, function (i) {
x <- st_sample (i$geometry,
size = addressesPerPostcode)
size = i$stroke_rate)
st_sf (POSTCODE = i$Postcode,
geometry = x)
})
randomPoints <- do.call (rbind, randomPoints)
st_crs (randomPoints) <- 4326
randomPoints_py <- apply (basicDemographicsRehab_py, 1, function (i) {
x <- st_sample (i$geometry,
size = addressesPerPostcode)
size = i$stroke_rate)
st_sf (POSTCODE = i$Postcode,
geometry = x)
})
Expand Down
89 changes: 49 additions & 40 deletions RehabCatchment/rvspy/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -86,17 +86,6 @@ basicDemographicsRehab <- filter(basicDemographicsVIC,
select(-starts_with("POA_"))
```

The python code has fewer postcodes than the R code, with numbers
determined manually here by comparing the corresponding maps. The
reduced version equivalent to the python code is:

``` r
#mapview::mapview (basicDemographicsRehab_py)
removes <- c (40, 56, 57, 53, 43, 10, 7, 8, 29, 11, 1, 3)
index <- seq (nrow (basicDemographicsRehab))
basicDemographicsRehab_py <- basicDemographicsRehab [!index %in% removes, ]
```

## Data sampling

The major difference between the R and python code is the sampling
Expand All @@ -107,8 +96,28 @@ points from within the postcode polygons. These two approaches are
replicated here in R code, the first referred to as `randomaddresses`,
the second as `randomPoints`.

The `addressesPerPostcode` value below is modified by the estimated
stroke rate per postcode calculated in the python code.

``` r
addressesPerPostcode <- 1000
addressesPerPostcode <- 10
stroke_rate <- read.csv ("../../python/notebooks/data/postcode_strokes.csv",
stringsAsFactors = FALSE)
stroke_rate$POSTCODE <- substr (stroke_rate$POA_CODE, 4, 7)
stroke_rate$strokes <- round (stroke_rate$strokes * addressesPerPostcode)
index <- match (basicDemographicsRehab$Postcode, stroke_rate$POSTCODE)
basicDemographicsRehab$stroke_rate <- stroke_rate$strokes [index]
```

The python code has fewer postcodes than the R code, with numbers
determined manually here by comparing the corresponding maps. The
reduced version equivalent to the python code is:

``` r
#mapview::mapview (basicDemographicsRehab_py)
removes <- c (40, 56, 57, 53, 43, 10, 7, 8, 29, 11, 1, 3)
index <- seq (nrow (basicDemographicsRehab))
basicDemographicsRehab_py <- basicDemographicsRehab [!index %in% removes, ]
```

Random addresses:
Expand All @@ -122,14 +131,14 @@ samplePCode <- function(pcode, number) {

randomaddresses <- map(basicDemographicsRehab$Postcode,
samplePCode,
number=addressesPerPostcode) %>%
number=basicDemographicsRehab$stroke_rate) %>%
bind_rows() %>%
sf::st_as_sf(coords = c("LONGITUDE", "LATITUDE"),
crs=st_crs(basicDemographicsRehab),
agr = "constant")
randomaddresses_py <- map(basicDemographicsRehab_py$Postcode,
samplePCode,
number=addressesPerPostcode) %>%
number=basicDemographicsRehab$stroke_rate) %>%
bind_rows() %>%
sf::st_as_sf(coords = c("LONGITUDE", "LATITUDE"),
crs=st_crs(basicDemographicsRehab),
Expand All @@ -141,15 +150,15 @@ Random points:
``` r
randomPoints <- apply (basicDemographicsRehab, 1, function (i) {
x <- st_sample (i$geometry,
size = addressesPerPostcode)
size = i$stroke_rate)
st_sf (POSTCODE = i$Postcode,
geometry = x)
})
randomPoints <- do.call (rbind, randomPoints)
st_crs (randomPoints) <- 4326
randomPoints_py <- apply (basicDemographicsRehab_py, 1, function (i) {
x <- st_sample (i$geometry,
size = addressesPerPostcode)
size = i$stroke_rate)
st_sf (POSTCODE = i$Postcode,
geometry = x)
})
Expand Down Expand Up @@ -256,39 +265,39 @@ kable (cases_per_centre (randomaddresses, net, nodes, RehabLocations))

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 10809 | 19.42807 |
| DandenongHospital | 16355 | 29.39643 |
| KingstonHospital | 28472 | 51.17550 |
| CaseyHospital | 21 | 19.09091 |
| DandenongHospital | 32 | 29.09091 |
| KingstonHospital | 57 | 51.81818 |

``` r
kable (cases_per_centre (randomaddresses, net_unwt, nodes, RehabLocations))
```

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 11231 | 20.18657 |
| DandenongHospital | 15817 | 28.42943 |
| KingstonHospital | 28588 | 51.38400 |
| CaseyHospital | 22 | 20.00000 |
| DandenongHospital | 31 | 28.18182 |
| KingstonHospital | 57 | 51.81818 |

``` r
kable (cases_per_centre (randomPoints, net, nodes, RehabLocations))
```

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 10806 | 19.17555 |
| DandenongHospital | 16333 | 28.98337 |
| KingstonHospital | 29214 | 51.84107 |
| CaseyHospital | 5699 | 15.51213 |
| DandenongHospital | 12708 | 34.58995 |
| KingstonHospital | 18332 | 49.89793 |

``` r
kable (cases_per_centre (randomPoints, net_unwt, nodes, RehabLocations))
```

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 11198 | 19.87117 |
| DandenongHospital | 15924 | 28.25759 |
| KingstonHospital | 29231 | 51.87124 |
| CaseyHospital | 5859 | 15.94763 |
| DandenongHospital | 12423 | 33.81420 |
| KingstonHospital | 18457 | 50.23817 |

``` r

Expand All @@ -298,39 +307,39 @@ kable (cases_per_centre (randomaddresses_py, net, nodes, RehabLocations))

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 6066 | 13.79671 |
| DandenongHospital | 14202 | 32.30150 |
| KingstonHospital | 23699 | 53.90179 |
| CaseyHospital | 13 | 14.77273 |
| DandenongHospital | 26 | 29.54545 |
| KingstonHospital | 49 | 55.68182 |

``` r
kable (cases_per_centre (randomaddresses_py, net_unwt, nodes, RehabLocations))
```

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 6464 | 14.70194 |
| DandenongHospital | 13675 | 31.10287 |
| KingstonHospital | 23828 | 54.19519 |
| CaseyHospital | 13 | 14.77273 |
| DandenongHospital | 26 | 29.54545 |
| KingstonHospital | 49 | 55.68182 |

``` r
kable (cases_per_centre (randomPoints_py, net, nodes, RehabLocations))
```

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 6139 | 13.69518 |
| DandenongHospital | 14031 | 31.30103 |
| KingstonHospital | 24656 | 55.00379 |
| CaseyHospital | 2952 | 10.95402 |
| DandenongHospital | 8688 | 32.23867 |
| KingstonHospital | 15309 | 56.80730 |

``` r
kable (cases_per_centre (randomPoints_py, net_unwt, nodes, RehabLocations))
```

| Destination | total | percent |
| :---------------- | ----: | -------: |
| CaseyHospital | 6533 | 14.57413 |
| DandenongHospital | 13578 | 30.29046 |
| KingstonHospital | 24715 | 55.13541 |
| CaseyHospital | 3090 | 11.46610 |
| DandenongHospital | 8518 | 31.60785 |
| KingstonHospital | 15341 | 56.92605 |

And that only makes a very small difference, in spite of the huge
apparent difference in distributions of random points, and still does
Expand Down

0 comments on commit 35c98ca

Please sign in to comment.