WeatherPrediction

WeatherPrediction is a simple set of tools for predicting weather using machine learning.
It uses TensorFlow, pandas, joblib, NumPy, and scikit-learn.

By default, the project is configured to work with the Meteostat database.

Installation

Make sure you have the required packages installed:

pip install tensorflow pandas numpy scikit-learn joblib

Setup

1. Open load_data.py and set the filenames for the scalers:

joblib.dump(input_scaler, "../models/weather_model_4_input_scaler.pkl")
joblib.dump(target_scaler, "../models/weather_model_4_target_scaler.pkl")

2. Open model_training.py and set the training data path="../data/12375.csv"

_input, target, dates = get_data(path="../data/12375.csv", columns=columns,
                         time_star = "2018-05-21", time_stop = "2025-05-31", for_training=True)

3. Open model_training.py and set the model save path:

model.save("../models/weather_model_4.h5")

4. Set the number of training epochs and start training:

model.fit(_input, target, epochs=1000, validation_split=0.2)

You can also modify the model structure, but remember to update the corresponding parts in load_data.py.

Usage

After training the model, you can make predictions by running main.py.

1. Load the model and scalers:

model = load_model("models/weather_model_3.h5")
scaler_input = joblib.load("models/weather_model_3_input_scaler.pkl")
scaler_target = joblib.load("models/weather_model_3_target_scaler.pkl")

2. Load your input data (make sure the date range and format are correct and match load_data.py settings):

_input, target, dates = get_data(
   path="data/Libertow Weather History.csv",
   columns=columns,
   time_star="2024-06-13",
   time_stop="2025-06-10",
   i_scaler=scaler_input,
   t_scaler=scaler_target
)

Ensure your data has at least 361 days (default requirement).

4. Run main.py to get your weather predictions.

License

MIT