YODA - ML bike Barcelona y Santander integrados junto con Malaga Parking

• Barcelona: bike
  • https://api.bsmsa.eu/ext/api/bsm/gbfs/v2/en/station_status
  • https://api.bsmsa.eu/ext/api/bsm/gbfs/v2/en/station_status https://api.bsmsa.eu/ext/api/bsm/gbfs/v2/en/station_information
  • https://opendata-ajuntament.barcelona.cat/data/es/dataset/informacio-estacions-bicing
  • https://opendata-ajuntament.barcelona.cat/data/es/dataset/estat-estacions-bicing
• Santander: bike
  • http://datos.santander.es/resource/?ds=estado-estaciones-bicicletas&id=c798a2d2-bb27-452e-9a97-25d97b5c21ac&ft=JSON
  • http://datos.santander.es/api/rest/datasets/tusbic_puestos_libres.json?items=17
• Malaga: parking
  • https://datosabiertos.malaga.eu/dataset/ocupacion-aparcamientos-publicos-municipales
  • https://datosabiertos.malaga.eu/dataset/ubicacion-de-aparcamientos-publicos-municipales

Description

• Barcelona's endpoint returns the number of bikes available in the requested station
• Santander's endpoint returns the number of bikes available in the requested station
• Malaga's endpoint returns the percentage of occupancy

Initial tasks

• Copy the template.env file into .env and configure it.
• Create ckan_yoda network

docker network create ckan_yoda

Versions

• Spark 2.4.5
• Scala 2.11.12
• Java 8

Training

• Clone this project

• If the model already exists (prediction-job/model) just build de project packages

docker compose -f docker-compose.build.packages.yml up -d --build

• If the model does not exist: Build the packages and train the model

docker compose -f docker-compose.train.yml up -d --build

• View the spark deployment in http://localhost:4040/jobs/

• Test that the prediction-job/model folder is generated with the ML model

Testing (Development)

• Run the whole scenario in dev

docker compose -f docker-compose.dev.yml up -d

• Initialize the nifi recollection (Nifi will recollect data every 30 minutes and will store only the last 24 hours of historical data):

  • Enter in the nifi interface: https://localhost:9090/nifi

  • Log in with the following credentials:

    • username: root
    • password: pass1234567890

  • Upload the template stored in ./nifi/Barcelona_template.xml

  • Upload the template stored in ./nifi/Santander_template.xml

  • Run every component deployed by the template

• Create the prediction entities and the subscriptions:

  • Enter in the orion container

  docker exec -it orion /bin/bash

  • Create the predictionEntities and the subscriptions like in the entities folder (createPredictionEntities.sh and subscribeReqPrediction.sh).

Test the solution

1. Validate that the subscription and entities exist (inside the orion container):

curl --location --request GET 'http://localhost:1026/ngsi-ld/v1/subscriptions/'

curl --location --request GET 'http://localhost:1026/ngsi-ld/v1/entities/urn:ngsi-ld:ReqPrediction1'

curl --location --request GET 'http://localhost:1026/ngsi-ld/v1/entities/urn:ngsi-ld:ResPrediction1'

2. Update the ReqPrediction1

curl --location --request PATCH 'http://localhost:1026/ngsi-ld/v1/entities/urn:ngsi-ld:ReqPrediction1/attrs' \
--header 'Content-Type: application/json' \
--data-raw '{
   "idStation": {
      "type":"Property",
      "value": [VALUE from 1 to 17 in Santander or from 1 to 505 in Barcelona or from 0 to 9 in Malaga] as a String
   },
   "weekday":{
      "type":"Property",
      "value": [VALUE from 1 to 7]
   },
   "hour":{
      "type":"Property",
      "value": [VALUE from 0 to 23]
   },
   "month":{
       "type": "Property",
       "value": [VALUE from 1 to 12]
   },
   "city":{
       "type": "Property",
       "value": ["Santander" or "Barcelona" or "Malaga"]
   },
   "predictionId":{
      "type":"Property",
      "value":"p-1662768034900"
   },
   "socketId":{
      "type":"Property",
      "value":"Fn0kKHEF-dOcr311AAAF"
   }
}'

Being:

• idStation: station id. (In Malaga each number represents one station name, assigned by their position in the following array: ["Salitre", "Cervantes","El_Palo","Av._de_Andalucia","Camas","Cruz_De_Humilladero","Alcazaba","San_Juan_De_La_Cruz","Pz._de_la_Marina" or "Tejon_y_Rodriguez"])
• month: [1, 2, 3, ..., 12]
• weekday: [1, ..., 7] 1 ->Sunday 7->Saturday
• time: : [0, ... , 23]
• predictionId: String to identify the prediction in the consuming application
• socketId: String to identify the socket with the client in the consuming application

3. See if the ResPrediction1 changes

curl --location --request GET 'http://localhost:1026/ngsi-ld/v1/entities/urn:ngsi-ld:ResPrediction1'

Response:

{
   "@context":"https://uri.etsi.org/ngsi-ld/v1/ngsi-ld-core-context.jsonld",
   "id":"urn:ngsi-ld:BarcelonaBikePrediction1",
   "type":"ResPrediction",
   "socketId":{
      "type":"Property",
      "value":"Fn0kKHEF-dOcr311AAAF"
   },
   "predictionId":{
      "type":"Property",
      "value":"p-1662768034900"
   },
   "predictionValue":{
      "type":"Property",
      "value": [PREDICTED VALUE]
   },
   "idStation":{
      "type":"Property",
      "value": [VALUE set at the request]
   },
   "weekday":{
      "type":"Property",
      "value": [VALUE set at the request]
   },
   "hour":{
      "type":"Property",
      "value": [VALUE set at the request]
   },
   "month":{
      "type":"Property",
      "value": [VALUE set at the request]
   },
   "city":{
      "type":"Property",
      "value": [VALUE set at the request]
   }

Production (YODA)

• Run the whole scenario in prod within YODA (only spark). You need to create the predictionEntities and the subscriptions like in the entities folder. They are required the creation of entities and the subscription of spark. In the consuming application there are two possibilities:
  • The application receives a notification when the prediction is made
  • The application asks periodically to orion (urn:ngsi-ld:ResPrediction1) and see if the prediction was made

docker compose up -d

• Example of petition made to ask for a prediction:

curl --location --request PATCH 'http://138.4.22.130/ngsi-ld/v1/entities/urn:ngsi-ld:ReqPrediction1/attrs' \
--header 'Content-Type: application/json' \
--data-raw '{...}'

Testing everything worked

1. Validate that the subscription and entities exist:

curl --location --request GET 'http://138.4.22.130/ngsi-ld/v1/subscriptions/'

curl --location --request GET 'http://138.4.22.130/ngsi-ld/v1/entities/urn:ngsi-ld:ReqPrediction1'

curl --location --request GET 'http://138.4.22.130/ngsi-ld/v1/entities/urn:ngsi-ld:ResPrediction1'

2. Update the ReqPrediction1 to the direction

curl --location --request PATCH 'http://138.4.22.130/ngsi-ld/v1/entities/urn:ngsi-ld:ReqPrediction1/attrs' \
--header 'Content-Type: application/json' \
--data-raw '{
   "idStation": {
      "type":"Property",
      "value": "1"
   },
   "weekday":{
      "type":"Property",
      "value": 3
   },
   "hour":{
      "type":"Property",
      "value": 20
   },
   "month":{
       "type": "Property",
       "value": 3
   },
   "city":{
       "type": "Property",
       "value": "Barcelona"
   },
   "predictionId":{
      "type":"Property",
      "value":"p-1662768034900"
   },
   "socketId":{
      "type":"Property",
      "value":"Fn0kKHEF-dOcr311AAAF"
   }
}'

3. See if the ResPrediction1 changes

curl --location --request GET 'http://138.4.22.130/ngsi-ld/v1/entities/urn:ngsi-ld:ResPrediction1'

Response:

{
    "@context": "https://uri.etsi.org/ngsi-ld/v1/ngsi-ld-core-context.jsonld",
    "id": "urn:ngsi-ld:ResPrediction1",
    "type": "ResPrediction",
    "socketId": {
        "type": "Property",
        "value": "Fn0kKHEF-dOcr311AAAF"
    },
    "predictionId": {
        "type": "Property",
        "value": "p-1662768034900"
    },
    "predictionValue": {
        "type": "Property",
        "value": 10
    },
    "idStation": {
        "type": "Property",
        "value": "1"
    },
    "weekday": {
        "type": "Property",
        "value": 3
    },
    "hour": {
        "type": "Property",
        "value": 20
    },
    "month": {
        "type": "Property",
        "value": 3
    },
    "city": {
        "type": "Property",
        "value": "Barcelona"
    }
}