Machine learning tools

Warning

This part of CliMetLab is still a work in progress. Documentation and code behaviour will change.

Todo

TODO: Develop and document machine learning related tools

to_tfdataset()

To use a CliMetLab dataset with tensorflow, use the _to_tfdataset() method.

>>> import climetlab as cml
>>> ds = cml.load_dataset(x)
>>> x = ds.to_tfdataset(options)
>>> model.fit(x, ....)

The discussion is still open to decide whether to_dataset() returns:

  • tf.keras.utils.experimental.DatasetCreator

  • tf.data.Dataset

  • tf.keras.utils.Sequence

  • A custom CliMetLab class

PyTorch support

Todo

A long-term goal of CliMetLab is to provide a easy way to use a dataset with Pytorch. A merge request within this respect would be welcome. The CliMetLab API (exposed to the end-user) should be mostly identical for Tensorflow or Pytorch.