ml4xcube 0.0.5

ML package for data cubes

ML-Toolkits

The ML Toolkits provide a set of best practice Python-based Jupyter Notebooks that showcase the implementation of the three start-of-the-art Machine Learning libraries (1) scikit-learn, (2) PyTorch and (3) TensorFlow based on the Earth System Data Cube.

Installation

You can install ml4xcube directly via pip:

pip install ml4xcube

Make sure you have Python version 3.8 or higher.

If you're planning to use ml4xcube with TensorFlow or PyTorch, set up these frameworks properly in your Conda environment.

Features

• Data preprocessing and normalization/standardization functions
• Gap filling features
• Dataset creation and train-/ test split sampling techniques
• Trainer classes for sklearn, TensorFlow and PyTorch
• Distributed training framework compatible with PyTorch
• chunk utilities for working with data cubes

Usage

To use ml4xcube in your project, simply import the necessary module:

from ml4xcube.statistics import normalize, standardize
from ml4xcube.training.pytorch import Trainer
# Other imports...

You can then call the functions directly:

# Normalizing data
normalized_data = normalize(your_data, data_min, data_max)

# Trainer instance
trainer = Trainer(
    model           = reg_model,
    train_data      = train_loader,
    test_data       = test_loader,
    optimizer       = optimizer,
    best_model_path = best_model_path,
    early_stopping  = True,
    patience        = 3,
    epochs          = epochs
)

# Start training
reg_model = trainer.train()

License

ml4xcube is released under the MIT License. See the LICENSE file for more details.