Train/Valid/Test Split

[Hannes Bradl]

Hi,

I would like to train a model on the whole AudioSet. However, I'm not sure how to split the dataset into training, validation and test set.

At the moment I'm using the unbalanced train segments for training, the balanced train segments for validation and the eval segments for testing.

Is this a reasonable approach?

Best regards,

Hannes