Part 3: A simple problem with 3 foods and 3 constraints

We document the optimization procedure with nloptr using a small example of 3 foods, and 3 constraints (energy, protein, ghge).

foods <- read.csv('data/foods.csv', sep = ',')
data.table::setDT(foods) # use data.table format

# we only take 3 foods
fd <- foods[food %in% c('Bread', 'Vegetables', 'Red meat')]
fd
         food intake energy protein   fat carbs sugar alcohol  ghge
1:      Bread  175.4 10.696   0.091 0.030 0.441 0.002       0 0.001
2: Vegetables  154.6  1.565   0.015 0.008 0.050 0.005       0 0.001
3:   Red meat  117.6  8.342   0.173 0.139 0.014 0.000       0 0.013

Constraints for 3 foods

We compute the constraints for 3 foods together. Note that the consumption for each of the 28 food groups are different, hence we can not use 3/28 times the total energy, protein…; we need to use a weighted average.

The cosntraints can be set to any positive number that make sense.

# original constraint for all foods (maximum)
const_max_allfoods <- c(9314.3, 98.2, 85.8, 234.7, 39.2, 8.6, 3.8)

# 3 foods contribution (maximum)
const_max_3foods <- t(as.matrix(fd$intake)) %*% as.matrix(fd[, .(energy, protein, fat, carbs, sugar, alcohol, ghge)])
const_max_3foods
       energy protein     fat   carbs  sugar alcohol   ghge
[1,] 3099.047 38.6252 22.8452 86.7278 1.1238       0 1.8588

In the optimization problem, we want to have a range (lower, upper) for the metrics. We can assume that the lower range is 90% of the maximum.

c3foods <- rbind(const_max_3foods*0.9, const_max_3foods*1)
rownames(c3foods) <- c('lwr', 'upr')
c3foods <- data.frame(c3foods)
c3foods
      energy  protein      fat    carbs   sugar alcohol    ghge
lwr 2789.142 34.76268 20.56068 78.05502 1.01142       0 1.67292
upr 3099.047 38.62520 22.84520 86.72780 1.12380       0 1.85880

For example, the total energy for the 3 foods together should be within [2789.14, 3099.05].

Optimization

Now we are going to solve the optimization problem using nloptr.

There are a few components that need to be specified:

  • objective function (as a function of x)
  • constraint functions
  • initial values
  • options

Objective function

The objective is to minimize the following function:

\((x_1 - X_1)^2 + (x_2 - X_2)^2 + (x_3 - X_3)^2\) where

\(x_1, x_2, x_3\) are the target diet (grams or other units) we want to find, these three are unknown.

\(X_1, X_2, X_3\) are the current diet for bread, vegetables and red meat. We have these values in the data: 175.4, 154.6, 117.6. In the function we use fd$intake to programmatically extract the values.

fd$intake
[1] 175.4 154.6 117.6
# define objective 
objective <- function(x)
{
  return ( (x[1]- fd$intake[1])^2 + 
             (x[2]- fd$intake[2])^2 + 
             (x[3]- fd$intake[3])^2)
}

Constraints

Now we define the inequality constraints. For this demo, we only include 3 metrics (energy, protein, ghge) for simplicity.

# select the metrics we want to keep 
c3 <- c3foods[, c('energy', 'protein', 'ghge')]
c3
      energy  protein    ghge
lwr 2789.142 34.76268 1.67292
upr 3099.047 38.62520 1.85880

The inequality constraints need to be reformulated for nloptr to work. More specifically, nloptr optimizes a function \(f(x)\) subject to a set of equality and inequality constraints. For the inequality constraints, they need to be the form of \(g(x) <= 0\).

We write out our own constraints for energy:

\(x_1 e_1 + x_2e_2 + x_3e_3 >= E_{lower}, x_1 e_1 + x_2e_2 + x_3e_3 <= E_{upper}\)

These two need to be re-written to be in the format of \(g(x) <= 0\).

\(-(x_1 e_1 + x_2e_2 + x_3e_3)+E_{lower} <= 0\)

\(x_1 e_1 + x_2e_2 + x_3e_3 - E_{upper} <= 0\)

The constraints for protein, ghge can be written in similar ways.

# define the inequality constraints
inequalconstr <- function (x) {
  constr <- c(
    # energy
    - x[1]*fd$energy[1] - x[2]*fd$energy[2] - x[3]*fd$energy[3] + c3$energy[1], # lower
    x[1]*fd$energy[1] + x[2]*fd$energy[2] + x[3]*fd$energy[3] - c3$energy[2], # upper
    
    # protein
    - x[1]*fd$protein[1] - x[2]*fd$protein[2] - x[3]*fd$protein[3] + c3$protein[1],
    x[1]*fd$protein[1] + x[2]*fd$protein[2] + x[3]*fd$protein[3] - c3$protein[2],

    # ghge
    - x[1]*fd$ghge[1] - x[2]*fd$ghge[2] - x[3]*fd$ghge[3]+ c3$ghge[1],
    x[1]*fd$ghge[1] + x[2]*fd$ghge[2] + x[3]*fd$ghge[3] - c3$ghge[2]
  )
  return (constr)
}

Other options

We do not need to explicitly specify the non-negativity constraints for \(x\) inside the function. They are specified outside the function as a set of lower and upper bounds. You can set these range to be anything that make sense; for now we make it close to the current diet.

Initial values are the values you tell the program to start the search. This is also up to you; sometimes initial values can have a large impact on the final results when the optimization problem is difficult (e.g. has multiple local optima). We can start close to the values of the current diet.

# lower and upper bounds of x (3 foods)
lb <- c(160, 140, 100)
ub <- c(180, 160, 120)

# Initial values
# (try different ones!)
x0 <- c(175, 150, 110)

Options should also be supplied to the program as a named list. These are specifications for the algorithm, tolerance and maximum evaluation. For more details please check the function manual.

Pay attention to the last line here: the size inside rep() need to match the number of inequality constraints. In this example it should be 6.

opts <- list( "algorithm" = "NLOPT_GN_ISRES",
              "xtol_rel"= 1.0e-15,
              "maxeval"= 160000,
              "tol_constraints_ineq" = rep( 1.0e-10, 6 ))

Now we run the optimization by calling nloptr from nloptr package. If you have loaded the package before, you don’t need the :: between the two.

We print the result and check the optimized values.

# run the algorithm
res <- nloptr::nloptr(
  x0          = x0,        # initial value for x
  eval_f      = objective, # objective function
  lb          = lb,        # lower bound for x
  ub          = ub,        # upper bound for x
  eval_g_ineq = inequalconstr, # inequality constraint
  opts        = opts       # options
  )

print(res)

Call:
nloptr::nloptr(x0 = x0, eval_f = objective, lb = lb, ub = ub, 
    eval_g_ineq = inequalconstr, opts = opts)


Minimization using NLopt version 2.7.1 

NLopt solver status: 4 ( NLOPT_XTOL_REACHED: Optimization stopped because 
xtol_rel or xtol_abs (above) was reached. )

Number of Iterations....: 35522 
Termination conditions:  xtol_rel: 1e-15    maxeval: 160000 
Number of inequality constraints:  6 
Number of equality constraints:    0 
Optimal value of objective function:  9.22904150236166e-26 
Optimal value of controls: 175.4 154.6 117.6

The optimal values are 175.4, 154.6, 117.6. These are the same as the current diet.

Note

This result is not surprising; our current diet on the three foods already satisfies the constraints, and it is the closest to the current diet (\((x_1 - X_1)^2 +(x_2 - X_2)^2 + (x_3 - X_3)^2 = 0\)) when \(x_1 = X_1, x_2 = X_2, x_3 = X_3\).

Reduce ghge to 80%

We can try to modify the constraint values to see if the diet will be different.

For example, we can limit ghge. The current upper and lower range are [1.673, 1.859], let us try to limit it to [1.338, 1.487], which is 80% of the original values.

# c3foods <- rbind(const_max_3foods*0.9, const_max_3foods*1)
# rownames(c3foods) <- c('lwr', 'upr')
# c3foods <- data.frame(c3foods)
c3
      energy  protein    ghge
lwr 2789.142 34.76268 1.67292
upr 3099.047 38.62520 1.85880
c3_alt <- c3
c3_alt$ghge <- c3_alt$ghge * 0.8 # you can try different limits
c3_alt
      energy  protein     ghge
lwr 2789.142 34.76268 1.338336
upr 3099.047 38.62520 1.487040

We keep the objective function and options same as before; but we modify the inequality constraints.

# define the inequality constraints
# instead of c3, we use c3_alt

inequalconstr_alt <- function (x) {
  constr <- c(
    # energy
    - x[1]*fd$energy[1] - x[2]*fd$energy[2] - x[3]*fd$energy[3] + c3_alt$energy[1], # lower
    x[1]*fd$energy[1] + x[2]*fd$energy[2] + x[3]*fd$energy[3] - c3_alt$energy[2], # upper
    
    # protein
    - x[1]*fd$protein[1] - x[2]*fd$protein[2] - x[3]*fd$protein[3] + c3_alt$protein[1],
    x[1]*fd$protein[1] + x[2]*fd$protein[2] + x[3]*fd$protein[3] - c3_alt$protein[2],

    # ghge
    - x[1]*fd$ghge[1] - x[2]*fd$ghge[2] - x[3]*fd$ghge[3]+ c3_alt$ghge[1], # new values
    x[1]*fd$ghge[1] + x[2]*fd$ghge[2] + x[3]*fd$ghge[3] - c3_alt$ghge[2] # new values
  )
  return (constr)
}

Now we run the optimization again with the new constraint, inequalconstr_alt.

# run the algorithm
res_alt <- nloptr::nloptr(
  x0          = x0,        # initial value for x
  eval_f      = objective, # objective function
  lb          = lb,        # lower bound for x
  ub          = ub,        # upper bound for x
  eval_g_ineq = inequalconstr_alt, # inequality constraint (NEW)
  opts        = opts       # options
  )

print(res_alt)

Call:
nloptr::nloptr(x0 = x0, eval_f = objective, lb = lb, ub = ub, 
    eval_g_ineq = inequalconstr_alt, opts = opts)


Minimization using NLopt version 2.7.1 

NLopt solver status: 5 ( NLOPT_MAXEVAL_REACHED: Optimization stopped because 
maxeval (above) was reached. )

Number of Iterations....: 160000 
Termination conditions:  xtol_rel: 1e-15    maxeval: 160000 
Number of inequality constraints:  6 
Number of equality constraints:    0 
Current value of objective function:  566.401428410334 
Current value of controls: 168.806 140 100

The new results are 168.806, 140, 100.

Compare the two runs
  • Bread: 175.4 -> 168.8 (3.7% reduction)
  • Vegetables: 154.6 -> 140 (9.4% reduction)
  • Red meat: 117.6 -> 100 (15.0% reduction)

This makes sense, as red meat is the largest contributor for ghge.

What we have shown here is a very crude demonstration of how to construct the optimization problem, and a sanity check of the results. A lot more can be tried out.