This repository contains code to instantiate and deploy a weather forecasting model. The 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 wind speed).
Three models have been included with this repository, 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. You can specify which model you wish to use when making requests to the API (see Use the Model below for more details).
A description of the weather variables used to train the models is set out below.
| Variable | Description |
|---|---|
| HOURLYVISIBILITY | Distance from which an object can be seen. |
| HOURLYDRYBULBTEMPF | Dry bulb temperature (degrees Fahrenheit). Most commonly reported standard temperature. |
| HOURLYWETBULBTEMPF | Wet bulb temperature (degrees Fahrenheit). |
| HOURLYDewPointTempF | Dew point temperature (degrees Fahrenheit). |
| HOURLYRelativeHumidity | Relative humidity (percent). |
| HOURLYWindSpeed | Wind speed (miles per hour). |
| HOURLYStationPressure | Atmospheric pressure (inches of Mercury; or 'in Hg'). |
| HOURLYSeaLevelPressure | Sea level pressure (in Hg). |
| HOURLYPrecip | Total precipitation in the past hour (in inches). |
| HOURLYAltimeterSetting | Atmospheric pressure reduced to sea level using temperature profile of the “standard” atmosphere (in Hg). |
| HOURLYWindDirectionSin | Sine component of wind direction transformation (since wind direction is cyclical). |
| HOURLYWindDirectionCos | Cosine component of wind direction transformation (since wind direction is cyclical). |
| HOURLYPressureTendencyIncr | Dummy variable indicating if pressure was increasing in the past hour. |
| HOURLYPressureTendencyDecr | Dummy variable indicating if pressure was decreasing in the past hour. |
| HOURLYPressureTendencyCons | Dummy variable indicating if pressure has stayed relatively constant in the past hour. |
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 15 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
The model files are provided as part of this repository in the assets/models folder. The code in this
repository deploys the model as a web service in a Docker container. This repository was developed as part of the
IBM Code Model Asset Exchange and the public API is powered by
IBM Cloud.
| Domain | Application | Industry | Framework | Training Data | Input Data Format |
|---|---|---|---|---|---|
| Weather | Time Series Prediction | General | TensorFlow / Keras | JFK Airport Weather Data, NOAA | CSV |
- Data from US Local Climatological Data, National Climatic Data Center, National Oceanic & Atmospheric Administration
Literature and Documentation
- LSTMs in Keras
- Time Series Prediction with RNNs
- S. Hochreiter, J. Schmidhuber "Long Short Term Memory", Neural Computation 1997
Related Repositories
| Component | License | Link |
|---|---|---|
| This repository | Apache 2.0 | LICENSE |
| Model Weights | Apache 2.0 | LICENSE |
| Test Assets | No restriction | Asset README |
docker: The Docker command-line interface. Follow the installation instructions for your system.- The minimum recommended resources for this model is 2GB Memory and 2 CPUs.
- If you are on x86-64/AMD64, your CPU must support AVX at the minimum.
To run the docker image, which automatically starts the model serving API, run:
$ docker run -it -p 5000:5000 quay.io/codait/max-weather-forecaster
This will pull a pre-built image from the Quay.io container registry (or use an existing image if already cached locally) and run it. If you'd rather checkout and build the model locally you can follow the run locally steps below.
You can deploy the model-serving microservice on Red Hat OpenShift by following the instructions for the OpenShift web console or the OpenShift Container Platform CLI in this tutorial, specifying quay.io/codait/max-weather-forecaster as the image name.
You can also deploy the model on Kubernetes using the latest docker image on Quay.
On your Kubernetes cluster, run the following commands:
$ kubectl apply -f https://github.com/IBM/MAX-Weather-Forecaster/raw/master/max-weather-forecaster.yaml
The model will be available internally at port 5000, but can also be accessed externally through the NodePort.
Clone this repository locally. In a terminal, run the following command:
$ git clone https://github.com/IBM/MAX-Weather-Forecaster.git
Change directory into the repository base folder:
$ cd MAX-Weather-Forecaster
To build the docker image locally, run:
$ docker build -t max-weather-forecaster .
Note that currently this docker image is CPU only (we will add support for GPU images later).
To run the docker image, which automatically starts the model serving API, run:
$ docker run -it -p 5000:5000 max-weather-forecaster
The API server automatically generates an interactive Swagger documentation page. Go to http://localhost:5000 to load it. From there you can explore the API and also create test requests.
Use the model/predict endpoint to load a test data file and get predictions for the relevant weather target variable (or variables) from the API. You can use one of the test files from the assets/lstm_weather_test_data folder, after unzipping the test data archive by running the following command:
$ tar -zxvf assets/lstm_weather_test_data.tar.gz -C assets
You can also test it on the command line, for example to test the univariate model:
$ curl -F "file=@assets/lstm_weather_test_data/univariate_model_test_data.txt" -XPOST http://localhost:5000/model/predictYou can select one of the three available models used to make predictions by setting the model request parameter to one of: univariate (default), multivariate, or multistep. Note that each model takes in different weather datasets. After loading a particular model, you must predict only on the accompanying test dataset (e.g. univariate must predict on univariate_model_test_data.txt).
For example, to test the multivariate model:
$ curl -F "file=@assets/lstm_weather_test_data/multivariate_model_test_data.txt" -XPOST http://localhost:5000/model/predict?model=multivariateTo test the multi-step model:
$ curl -F "file=@assets/lstm_weather_test_data/multistep_model_test_data.txt" -XPOST http://localhost:5000/model/predict?model=multistepYou 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
],
...
}
To run the Flask API app in debug mode, edit config.py to set DEBUG = True under the application settings. You will
then need to rebuild the Docker image (see step 1).
To stop the Docker container, type CTRL + C in your terminal.
If you are interested in contributing to the Model Asset Exchange project or have any queries, please follow the instructions here.