BalancingGroups
BalancingGroups copied to clipboard
facebookresearch
Simple data balancing baselines for worst-group-accuracy benchmarks.
Inside the Python script ***models.py***, there is a class named **GroupDRO** as the piece of the code given below, ``` class GroupDRO(ERM): def __init__(self, hparams, dataset): super(GroupDRO, self).__init__(hparams, dataset) self.register_buffer(...
With CivilComments, the methods SUBG and RWG use a different preprocessed version of the data (fine/coarse). The paper does not mention this, and only mention the "coarse" version (with "aggregated"...
Hi! I noticed that in your code for BERT AdamW optimizer you only apply weight decay to parameters that contain the strings `bias` or `LayerNorm.weight`: https://github.com/facebookresearch/BalancingGroups/blob/72d31e56e168b8ab03348810d4c5bac0f8a90a7a/models.py#L41-L45 The original group DRO...
Metadata
Owner
Metadata
Simple data balancing baselines for worst-group-accuracy benchmarks.