HURAIM

Hurricane Artificial Intelligence using Machine Learning

https://ams.confex.com/ams/101ANNUAL/meetingapp.cgi/Paper/380154

@article{Patel_Usmani_Brant_2021,
	title        = {Multivariate LSTM approach to hurricane intensity and tracking predictions},
	author       = {Patel, Akash B. and Usmani, Hammad and Brant, Jonathan C.},
	year         = 2021,
	month        = {Jan},
	journal      = {AMS},
	publisher    = {AMS},
	url          = {https://ams.confex.com/ams/101ANNUAL/meetingapp.cgi/Paper/380154}
}

Quickstart

Model Creation Command Line Arguments

Create the model specified with the command line. e.g.

>>> python run.py --universal
>>> python run.py --singular

Accepts command line argument as either, universal Creates a universal model with wind intensity, lat, and long singular Creates singular models with 3 different models for wind, lat and long If none are specified, we create a universal model. The following are examples of logs that are output on the command line.

[2023-11-04T21:51:24.100109Z] [HURAIM] Creating our training and validation data structure
[2023-11-04T21:51:24.108655Z] [HURAIM] Create universal features
[2023-11-04T21:51:24.110969Z] [HURAIM] Train for wind intensity (index 0), lat (index 1), lon (index 2).
[2023-11-04T21:51:24.192352Z] [HURAIM] Create and train bidirectional LSTM wind model

Training Command Line Arguments

--load

If there are models in the ml/models directory, we will use the files and weights in them according to the mode

    >>> python run.py                       # trains the artificial intelligence
    >>> python run.py --load                # loads the universal model weights
    >>> python run.py --singular --load     # loads the singular model weights

--epochs [int]

The number of epochs to train the model

    >>> python run.py --singular --epochs 100

Tutorial

python run.py -h
usage: run.py [-h] [--singular] [--universal] [--load] [--epochs EPOCHS] [--dropout DROPOUT] [--loss LOSS] [--optimizer OPTIMIZER]

optional arguments:
  -h, --help            show this help message and exit
  --singular            The 'singular' version of the architecture will be used
  --universal           The 'universal' version of the architecture will be used in ml/models
  --load                Loads existing model weights in the repository
  --epochs EPOCHS       Number of epochs to train the model
  --dropout DROPOUT     The dropout hyperparameter
  --loss LOSS           The loss hyperparameter
  --optimizer OPTIMIZER
                        The optimizer hyperparameter

Local Quickstart

Run this in the root directory HurricaneDissertaion.

pip install -r docker/files/requirements.txt

Git Quickstart

First, we open up a terminal and double check if we have git installed. Then, we can clone the code using this command and git. These commands are meant to be run inside the same directory. Currently, Github hosts our git repository.

git clone https://github.com/apatel726/HurricaneDissertation.git

When we make changes, we can see them here.

git diff

After confirming the changes, we can add them, or add all of the changes like here,

git add -A

Before uploading the changes, they all need a message pertaining to the development of the code.

git commit -m "Out of a mountain of despair, a stone of hope."

This command uploads the code to the main branch. If you are on a different branch, it is the same command to upload to the same branch on GitHub.

git push

MLFlow Quickstart

MLflow utilizes the port 5000 and we will need to have that open during our docker run such as,

docker run -it -p 8888:8888 -p 5000:5000 huraim

We can run the UI for the MLflow dashboard with this command,

mlflow ui

Please also reference https://www.mlflow.org/docs/latest/getting-started/quickstart-1/index.html

Docker Quickstart

Please note that this was taken from the docker/Docker.md file. Reference it for the most up to date documentation

Install Docker first. This Dockerfile should work on any OS. Please note that the following commands should be run inside the docker folder.

docker build -t huraim .
docker run --gpus all -it -p 8888:8888 -p 6006:6006 huraim

The above enables GPU. If GPU's are not available, replace with the following command,

docker run -it -p 8888:8888 -p 6006:6006 huraim

After running the above commands, open up a web browser and go to localhost:8888.

Creation Predictions Quickstart

Tutorial

Test File Prediction

python "test.py" --config "name of file to load configuration" --test "location of test file"

python test.py --config data/config.json --test data/hurdat2_test.txt

Live Hurricane Prediction

python "test.py" --config "name of file to load configuration" --test "location of test file"

python test.py --config data/config.json --test data/hurdat2_test.txt