It is a Semi-Supervised learning technique.
Assume that dataset has 2 parts: labeled data and un-labeled data. So how to use un-labeled data for training? => Using Pseudo Labelling.
Steps to train model using Pseudo Labelling:
- Train the model with labeled data.
- Use the model to predict un-labeled data.
- Choose the elements of un-labeled data that have predicted score greater than a threshold and number of chosen elements is calculated by (number of elements in unlabeled data * sample_rate).
- Append the chosen elements above into the train dataset and retrain the model.
Assume that dataset has 2 parts: labeled data and un-labeled data. So how to use un-labeled data for training? => Using Pseudo Labelling.
Steps to train model using Pseudo Labelling:
- Train the model with labeled data.
- Use the model to predict un-labeled data.
- Choose the elements of un-labeled data that have predicted score greater than a threshold and number of chosen elements is calculated by (number of elements in unlabeled data * sample_rate).
- Append the chosen elements above into the train dataset and retrain the model.
0 Comments