Unlabeled data preparation

[opencomvis]

Given a dataset that contains only 20 percent of annotated data divided into train/valid. How can we prepare the remaining 80% of the dataset that doesn't has labels ?