yeah i have checked
can we add MLP on top in custom Xlnetsequenceclassification
Xlnetsequenceclassification
already adds a linear
layer on top of XLNetModel
, but if you want to add an extra layer then yes, surely you can do it. You could use the XLNetForSequenceClassification
code for reference
sure thank you
I am using xlnet sequence classfication in dataset i have imbalance lables , i am facing overfit and only high count label is classified properly, what i can do?
You can set weight
in CrossEntropyLoss so that it can take the imbalance into account. You should fine enough information when you look this up. As an alternative, but in the same usage, you can use PyTorchâs weighted sampler.
thanks, how we can add class weights for xlnet sequence classification, any reference, please.
@BramVanroy
Hi i am working text classification, dataset contains imbalanced labels
"classcount = np.bincount(train_data[âlabelâ]).tolist()
train_weights = 1./torch.tensor(classcount, dtype=torch.float)
train__sampleweights = train_weights[train_data[âlabelâ]]
#class_weight
weighted_sampler = WeightedRandomSampler(
weights=train__sampleweights,
num_samples=len(train__sampleweights),
replacement=True)"
but still in validation its not predicted well for less labels