MET-Oracle is a tool to improve the lives of citizens and increase infrastructure's resilience, in Trinidad and Tobago, by helping us predict and prepare for adverse weather conditions brought on by Global Warming, and quantifying the effect it will have on our country's climate. It is a linear regression model built in Python using Tensorflow and our web interface uses the Google Maps API.
Met-Oracle is a regressor based on recurrent networks, using tensorflow.
Our objective is to predict continuous weather values, based on previous observations using the LSTM architecture.
This build is dependant on tensorflow-1.1.0 and the polyaxon library. (https://github.com/polyaxon/polyaxon)
A virtual env is recommended as this build is very dependant on the package versions listed in requirements.txt
python3
$ mkvirtualenv -p python3 ltsm
(ltsm) $
python2
$ mkvirtualenv ltsm
(ltsm) $
(ltsm) $ pip install -r ./requirements.txt
It was planned to use neural networks and supplied river levels to predict and help prepare for flooding, however, it was not feasible in the timeframe. In addition, Google Cloud Computing was the first choice for implementing the neural network so that users can directly interact with the model. For example, being able to enter a future date and then shown both curve plots and a geographical representation. Google Cloud has a free solution which we hoped and prepared to implement, however, our credit cards were refused during signup.