Project: Air quality data. Environmental and climate monitoring.

Atmosfer pressure. Regression model.

Artificial Intelligence  
Domain             : Regression
Sub-Domain         : Time series 
Architectures      : GradientBoostingRegressor
Application        : Meteological domain

Description of data

The system of automatic state ambient air monitoring stations in Lithuania consists of 14 urban air quality monitoring stations operating in Vilnius, Kaunas, Klaipėda, Šiauliai, Panevėžys, Jonava, Kėdainiai, Naujoji Akmenė and Mažeikiai and 3 integrated monitoring stations operating in Aukštaitija, Žemaitija and Dzūkija National Parks.

Concentrations of the following pollutants are measured at automatic air quality monitoring stations: particulate matter PM10, fine particulate matter PM2.5, nitrogen oxides (NO2, NOx, NO), sulfur dioxide (SO2), carbon monoxide (CO), ozone (O3), benzene, mercury. The tests and measurements shall be carried out in accordance with the requirements of Directives 2004/107/EC of the European Parliament and of the Council relating to arsenic, cadmium, mercury, nickel and polycyclic aromatic hydrocarbons in ambient air and 2008/50/EC on ambient air quality and cleaner air for Europe.

We will look at Atmosfer pressure measurements.

The data consists of files obtained from different sources(https://data.gov.lt/datasets/500/ ):

• Averages.csv - data air monitoring (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/Averages/:format/csv)
• Quantity.csv - dictionary (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/Quantity/:format/csv)
• QuantityUnits.csv - dictionary (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/QuantityUnits/:format/csv)
• Station.csv - dictionary (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/Station/:format/csv)

Metrics of the Final Model

Final Model             : model.bin
Final DictVectorizer    : dv.bin
Final MinMaxScaler      : scaler.bin
MAE:                    : Val - 3,  Test - 6 

Tools / Libraries

Languages               : Python
Tools/IDE               : Anaconda
Virtual Environment     : pipenv

Scripts and Notebook:

Notebook                : [https://github.com/Justlesia/predict-atm-pressure/blob/main/notebook.ipynb]
Predict Script          : [https://github.com/Justlesia/predict-atm-pressure/blob/main/predict.py]
Train Script            : [https://github.com/Justlesia/predict-atm-pressure/blob/main/train.py]
Test Script             : [https://github.com/Justlesia/predict-atm-pressure/blob/main/test_docker.py]
Test Script             : [https://github.com/Justlesia/predict-atm-pressure/blob/main/test_heroku.py]

Run the Model as is

Steps to run the scripts/notebooks as is:

1. Clone the repo by running the following command:

   git clone https://github.com/Justlesia/predict-atm-pressure.git
   

2. Download data into the ./datasets: (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/Averages/:format/csv) (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/Quantity/:format/csv) (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/QuantityUnits/:format/csv) (https://get.data.gov.lt/datasets/gov/aaa/oro_stociu_matavimai/Station/:format/csv)

Using Docker

1. Build and run the application using the commands:

   docker build -t predict . 
   docker run -it -p 9696:9696 predict:latest 
   

2. Open another terminal/prompt and run test_docker.py

   python test_docker.py
   

   If the prediction services gives prediction, then it means the docker container is working.

Deployment using heroku

Skip this step and go through to the 2 step.

1. Go to use heroku cml in the repository and type the following commands. The heroku.yml in the folder will be used to create a docker image and run the app.

   heroku login
   git init
   heroku git:remote -a name_of_the_project
   
   git add .
   git commit -m "add bild from heroku.yml"
   push heroku master
   

2. Now the app running at the address http://predict-atm-pressure-5bc65bbe7521.herokuapp.com/predict. To test if the docker container was built properly and running, go to the test_heroku.py

   python test_heroku.py
   

   If the prediction services give prediction, the heroku app is working.