Sharing my delivery system recipe

Features set

• geocoding: get GPS coordinate from address (text)
• directions: get itinerary from two addresses (text or GPS)
• dynamic pricing

Pricing

so far static

• area 1 : Centre-Ville 2.000 francs
• area 2 : Pikine, Guédiawaye, Parcelles 2.500 francs
• area 3 : Rufisque, Keur Massar 3.500 francs
• area 4 : Bargny, Diamniadio 5.000 Francs
• area x: price ?

Price = f(origin, destination)

The goal being to determine price per delivery. At first, I thought about having the pricing system to call the geocoding and direction services but separation of concerns principle. So from the addresses (origin, destination) -> geocoding for marker on map -> direction for itinerary -> pricing ()

What do you think the price for a delivery should depend on ? My answer to this question, is: distance and duration because traffic is something we do think about when taking a cab.

Price = f(distance, duration)

Data set 1

Direction service gave us

I am seeing this a linear function f(x,y) = ax + by + c

Numbers are in meters, seconds. I am taking only the average for the values above. That would give us

• 7600a + (13 * 60)b + c = 2000

• 14400a + (26 * 60 + 40)b + c = 2500

• 30250a + (43 * 60)b + c = 3500

• 36700a + (43 * 60)b + c = 5000

• (A) 7600a + 780b + c = 2000

• (B) 14400a + 1600b + c = 2500

• (C) 30250a + 2580b + c = 3500

• (D) 36700a + 2580b + c = 5000

B-A:6800a + 820b = 500 C-B:15850a + 980b = 1000

find a 6800a = 500 - 820b a = 1/6800(500 - 820b)

replace in C-B: 15850/6800(500 - 820b) + 980b = 1000 2.33(500 - 820b) + 980b = 1000 1165 - 930.6b = 1000 165 = 930.6b b = 0.1774

replace a a = 0.05215

7600a + 780b + c = 2000

c = 1465.37

a = 0.05215 b = 0.1774 c = 1465.37

price = 0.5215 * distance + 0.1774 * duration + 1465.37

Data set 2

Direction service gave us

• normalize data

const data = [
  [1000,'point E',2.1,7],
  [1000,'fann',2.1,5],
  [1000,'sacré coeur',2.3,9],
  [1000,'amitié',1.8,5],
  [1000,'castor',4.8,17],
  [1000,'dieuppeul',5.2,15],
  [1000,'HLM',4.1,12],
  [1000,'colobane',3.9,7],
  [1000,'fass',3.1,10],
  [1000,'grand dakar',2.8,9],
  [1000,'sicap liberté',4.4,11],
  [1500,'ouaka',3.3,8],
  [1500,'front de terre',5.5,16],
  [1500,'zone de captage',5.9 ,22],
  [1500,'liberté 6',4.1,9],
  [1500,'biscuiterie',4.2,15],
  [1500,'grand yoff',5.6,15],
  [2000,'Dakar plateau',7.6,13],
  [2000,'almadies',8.2,14],
  [2000,'mamelles',8.1,15],
  [2500,'Yarakh',4.8,16],
  [2500,'Dalifort',13.4,23],
  [2500,'Poste thiaroye',16.7,24],
  [2500,'Camberene',12.2,23],
  [3000,'Sicap mbao',27.2,39],
  [3000,'Grand mbao',26.9,39],
  [3000,'Guediawaye',17,30],
  [3000,'Pikine',15.7,29],
  [3000,'Thiaroye sur mer',15.7,27],
  [3000,'Guinaw rails',18.6,29],
  [3500,'Yeumbeul',20.4,41],
  [3500,'Keur massar',26.8,43],
  [3500,'Keur mbaye fall',23,35],
  [3500,'Malika',23.2,36],
  [3500,'Wakhinane nimzatt',18,30],
  [4000,'Rufisque',33.7,43],
  [4000,'Bargny',35.8,45],
  [4000,'Diamniadio',37.6,41],
  [4000,'Sebikhotane',50.3,67],
  [4000,'Sangalkam',36.7,48],
]

data.map(x => {
  return x[0] + ',' + x[2] + "\r\n"
})

const getColumnData = (data, columnIndex) => {
  return data.map(i => i[columnIndex])
}

const normalize = (columnData) => {
  const min = Math.min(...columnData)
  const max = Math.max(...columnData)
  return columnData.map(x => {
    return (x - min) / (max - min)
  })
}

const distances = getColumnData(data, 2)
const durations = getColumnData(data, 3)
normalize(distances)
normalize(durations)

| Price | Address | Distance (km) | Duration (minute) | 0.00000000e+00, 252.346741, -8.53235354e+00, 1.63897871e-01, -1.30360123e-03 price = (distance, duration) => 252.346741 * distance + -8.53235354 * duration - 0.00130360123

• Price per bracket

Retrospectives

it has been fun so far to have the system calculate it but it does need some level of human intervention:

• add smoothing for too small distance (ie 10m distance) or same point deliveries (ie dakar, dakar)
• allow user to move marker point

Wrap up

Maths are useful for a developer. Inference to determine pricing from a set of data