Training on Imbalanced Data

[mlnomadpy]

Hello,
Is there any implementation of class weighting in TensorFlow decision forests, similar to class_weight in Keras models?

[rstz]

Hi,
the preferred option is to set weights manually (e.g. give an explicit weight to each example). Doing so is possible by either adding a third channel to the tf.Dataset fed to TF-DF (see here) or (when converting from pandas) by specifying the weight parameter in pd_dataframe_to_tf_dataset, see here.

For binary classification, you can also set the class_weight parameter of the fit function. However, this is not supported for distributed training.

Finally the underlying C++ library Yggdrasil Decision Forests (YDF) provides more flexibility for defining weights through this proto message linked in the training configuration, but this is currently not fully exposed in TF-DF

[mlnomadpy]

Thank you very much for your response,
I will try to try the pd_dataframe_to_tf_dataset approach.