The nutrientprofiler package aims to provide a series of
R functions that implement UK Nutrient Profiling Model scoring that can
be applied across product datasets.
This is broken up into 3 parts within the package:
- Specific gravity conversions
- Scoring the different nutritional components of the NPM score
- Calculating the NPM score and NPM assessment logic
To start we load the package the with command below.
We can also load some example data to help us with the following examples. This data was created specifically for this package and so is potentially quite different to how your real world data might look.
dim(cdrcdrinks)
#> [1] 10 21
cdrcdrinks[1, ]
#> name brand product_category product_type food_type drink_format drink_type
#> 1 lembas NA NA Food
#> nutrition_info energy_measurement_kj energy_measurement_kcal
#> 1 266 NA
#> sugar_measurement_g satfat_measurement_g salt_measurement_g
#> 1 50 3 NA
#> sodium_measurement_mg fibre_measurement_nsp fibre_measurement_aoac
#> 1 0.6 3 NA
#> protein_measurement_g fvn_measurement_percent weight_g volume_ml
#> 1 7 0 100 NA
#> volume_water_ml
#> 1 NASpecific gravity conversions
When trying to determine the NPM score the volume or weight of the product needs to be adjusted to account for it’s specific gravity.
To adjust product weights or volumes for specific gravity we use the
SGConverter function. This is a high level function that is
designed to operate on each row of a data.frame parsing multiple columns
to determine how to calculate the adjusted specific gravity score.
Warning: The `SGConverter` function has been specifically designed with an
existing dataset in mind and expects specific column names to work.
# using dplyr
library(dplyr)
#>
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#>
#> filter, lag
#> The following objects are masked from 'package:base':
#>
#> intersect, setdiff, setequal, union
cdrcdrinks %>%
rowwise() %>%
mutate(out = SGConverter(pick(everything()))) %>%
select(out)
#> # A tibble: 10 × 1
#> # Rowwise:
#> out
#> <dbl>
#> 1 100
#> 2 130
#> 3 104
#> 4 129.
#> 5 103
#> 6 100
#> 7 51.5
#> 8 25
#> 9 124.
#> 10 109
# using base R
cdrcdrinks["sg"] <- unlist(lapply(seq_len(nrow(cdrcdrinks)), function(i) SGConverter(cdrcdrinks[i, ])))The below figure attempts to outline the hierarchy of function calls
that SGConverter initiates. The logic for determining how
to adjust values for specific gravity is complicated by the potential
options around whether a drink is ready-to-drink, a powdered
preparation, or a cordial, within both the powdered and cordial options
additional consideration must be given to the preparation instructions
that are provided. In the below figure each function is named in each
node and each column name the function uses to dispatch to underlying
functions is specified in single quotes. If a function returns a value
it is marked with an empty diamond.
SGConverter logic
Nutrient profile model scoring
The next part of this package is a series of functions for handling
Nutrient Profile Model scoring. These specifically look at functions for
adjusting units into the expected unit as documented for the Nutrient
Profile Model, a generic scoring function for returning a score for
a value given a number of thresholds, a scoring dispatcher function that
determines what adjuster and scoring thresholds to use for a given value
and a high level NPMScore wrapper function that shows the
logic for calculating the NPM score across multiple nutritional groups
for a data.frame of data.
A quick example of running NPMScore across a single row
of data is shown below:
NPMScore(cdrcdrinks[1,], sg_adjusted_label="sg")
#> energy_score sugar_score satfat_score protein_score salt_score fvn_score
#> 1 0 10 2 4 0 0
#> fibre_score
#> 1 4An example of building a tidyverse pipeline to calculate scores
across all rows of a data.frame using NPMScore and
SGConverter is shown below:
library(tidyr)
cdrcdrinks %>%
rowwise() %>%
mutate( sg = SGConverter(pick(everything()))) %>%
mutate(test = NPMScore(pick(everything()), sg_adjusted_label="sg")) %>%
unnest(test) %>% select(energy_score, sugar_score, salt_score, fvn_score,
protein_score, satfat_score, fibre_score)
#> # A tibble: 10 × 7
#> energy_score sugar_score salt_score fvn_score protein_score satfat_score
#> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 0 10 0 0 4 2
#> 2 0 3 0 0 1 8
#> 3 2 2 1 0 0 0
#> 4 0 2 0 0 0 0
#> 5 2 4 0 0 0 0
#> 6 0 4 0 0 2 10
#> 7 1 6 2 0 0 0
#> 8 2 10 4 0 1 0
#> 9 2 3 0 0 0 0
#> 10 2 3 0 0 0 0
#> # ℹ 1 more variable: fibre_score <dbl>Nutrient Profile Model assessment
After calculating the scores for specific nutrients for our product next we need to perform the actual Nutrient Profile Model assessment. This involves combining individual nutrient scores to calculate an A score and a C score and use these compound to calculate a total Nutrient Profile Model score and assess this to determine a pass or fail.
This package implements this logic in a high-level wrapper function
called NPMAssess which operates on a row of a data.frame.
It expects the columns generated in the previous NPMScore
step to allow it to calculate the A score and C score and can be used as
follows:
# create NPM_score data.frame from NPMScore
# using the specific gravity `sg` column created above
npm_scores <- do.call(
"rbind",
lapply(
seq_len(nrow(cdrcdrinks)),
function(i) NPMScore(cdrcdrinks[i, ], sg_adjusted_label = "sg")
)
)
# append NPM Score columns to original data
combo_df <- cbind(cdrcdrinks, npm_scores)
# test NPMAssess on this data
NPMAssess(combo_df[1, ])
#> A_score C_score NPM_score NPM_assessment
#> 1 12 8 8 FAILWe can also use tidyverse functions to build an entire pipeline for running Nutrient Profile Model assessments.
# using tidyr
library(tidyr)
cdrcdrinks %>%
rowwise() %>%
mutate( sg = SGConverter(pick(everything()))) %>%
mutate(test = NPMScore(pick(everything()), sg_adjusted_label="sg")) %>%
unnest(test) %>%
rowwise() %>%
mutate(assess = NPMAssess(pick(everything()))) %>%
unnest(assess) %>%
select(product_type, NPM_score, NPM_assessment)
#> # A tibble: 10 × 3
#> product_type NPM_score NPM_assessment
#> <chr> <dbl> <chr>
#> 1 Food 8 FAIL
#> 2 Food 11 FAIL
#> 3 Drink 5 FAIL
#> 4 Drink 2 FAIL
#> 5 Drink 6 FAIL
#> 6 Food 14 FAIL
#> 7 Drink 9 FAIL
#> 8 Drink 16 FAIL
#> 9 Drink 5 FAIL
#> 10 Drink 5 FAILThe building blocks of NPMAssess are explained in more
detail in the Nutrient
Profile Model Assessment vignette