Introduction
oasi-ozone-forecaster is a package for the forecast of the next day ozone maximum hourly concentration in Ticino, Switzerland. These forecasts are calculated twice: the night before for the upcoming day and the morning of the same day for the present day.
The forecasts are achieved through a ML approach which is trained on a set of both measured and forecasted meteorological values from 2015 to 2021. Please see the openly accessible published paper on the subject for more information.
Basic instructions
To properly configure the installation, refer to the following basic instructions:
Clone the repo:
git clone https://github.com/supsi-dacd-isaac/oasi-ozone-forecaster.gitMove to the newly created directory:
cd oasi-ozone-forecasterCreate
logsdirectory:mkdir logs/Create a virtual environment:
python3 -m venv venv/Install the necessary packages:
venv/bin/python3 -m pip install numpy
venv/bin/python3 -m pip install -r requirements.txt
Necessary adjustaments
The skgarden package is severely outdated, thus some manual modifications need to be applied in folder /venv/lib64/python3.x/site-packages/skgarden/:
In file
quantile/ensemble.py, line 40: addsample_weightto method argumentsIn file
quantile/ensemble.py, line 79: addsample_weightto method callIn file
quantile/tree.py, lines 221 and 232: removepresort=Falseoption to clean stdout output (optional)In file
mondrian/ensemble/forest.py, line 96 and 229: addsample_weightto method argumentsIn file
mondrian/ensemble/forest.py, line 121 and 254: addsample_weightto method call
Finally, in file classes/model_trainer.py uncomment line 11.
You are now ready to execute the scripts. Proceed to Main scripts and the descriptions therein.