Weather Forecaster

Overview

This model takes hourly weather data (as a Numpy array of various weather features, in text file format) as input and returns hourly weather predictions for a specific target variable or variables (such as temperature or windspeed).

Three pre-trained models are provided, all trained by the CODAIT team on National Oceanic and Atmospheric Administration local climatological data originally collected by JFK airport. All three models use an LSTM recurrent neural network architecture.

A description of the weather variables used to train the models is set out below.

For further details on the weather variables see the US Local Climatological Data Documentation

Each model returns a different format for its predictions:

• Univariate Model: returns a prediction of dry bulb temperature (HOURLYDRYBULBTEMPF), for the next hourly time step, for each input data point
• Multivariate Model: returns predictions for all 12 weather variables, for the next hourly time step, for each input data point
• Multistep Model: returns predictions of dry bulb temperature (HOURLYDRYBULBTEMPF), for the next 48 hourly time steps, for each input data point

Model Metadata

• Data from US Local Climatological Data, National Climatic Data Center, National Oceanic & Atmospheric Administration

References

Literature and Documentation

• LSTMs in Keras
• Time Series Prediction with RNNs
• S. Hochreiter, J. Schmidhuber “Long Short Term Memory”, Neural Computation 1997

Related Repositories

• TensorFlow Tutorials for Time Series
• Tensorflow LSTM Regression

Licenses

Options available for deploying this model

This model can be deployed using the following mechanisms:

• Deploy from Dockerhub:

  docker run -it -p 5000:5000 codait/max-weather-forecaster
  

  Show more

• Deploy on Red Hat OpenShift:

  Follow the instructions for the OpenShift web console or the OpenShift Container Platform CLI in this tutorial and specify codait/max-weather-forecaster as the image name.

• Deploy on Kubernetes:

  kubectl apply -f https://raw.githubusercontent.com/IBM/MAX-Weather-Forecaster/master/max-weather-forecaster.yaml
  

  Show more

  A more elaborate tutorial on how to deploy this MAX model to production on IBM Cloud can be found here.

• Locally: follow the instructions in the model README on GitHub

Example Usage

Once deployed, you can test the model from the command line. For example to test the multi-step model when running locally:

curl -F "file=@assets/lstm_weather_test_data/multistep_model_test_data.txt" -XPOST http://localhost:5000/model/predict?model=multistep

You should see a JSON response like that below for the multistep test data, where predictions contains the predicted dry bulb temperature (in F) for each of the next 48 hours, for each input data point.

{
  "status": "ok",
  "predictions": [
    [
      77.51201432943344,
      76.51381462812424,
      75.0168582201004,
      73.84445126354694,
      72.79087746143341,
      71.71804094314575,
      70.97693882882595,
      70.44060184061527,
      69.89843893051147,
      69.35454525053501,
      69.04163710772991,
      68.70432360470295,
      68.37075608968735,
      68.20421539247036,
      68.01852786540985,
      67.6653740555048,
      67.27566187083721,
      67.0398361980915,
      66.69407051801682,
      66.9289058893919,
      67.19844545423985,
      67.65162572264671,
      68.30480472743511,
      69.37090930342674,
      70.37226051092148,
      71.57235226035118,
      72.68855434656143,
      73.91224025189877,
      74.65138283371925,
      75.09161844849586,
      75.30447003245354,
      75.04770956933498,
      74.93723678588867,
      74.27759975194931,
      73.82458955049515,
      73.32358133792877,
      72.66812674701214,
      71.75925283133984,
      71.28871068358421,
      70.66486597061157,
      70.06835387647152,
      69.74887031316757,
      69.49707941710949,
      69.26406812667847,
      68.87126012146473,
      68.60496838390827,
      68.39429907500744,
      68.03596951067448
    ],
    ...
}

Resources and Contributions

If you are interested in contributing to the Model Asset Exchange project or have any queries, please follow the instructions here.