Meta Learning for POS Tagging

Implementation of three Meta Learning approaches : MAML, Reptile for POS Tagging in PyTorch.

Datset has been borrowed from Universal Dependencies

Nine languages are used to train the model.

Training the Model

This would also print the test results

to train and test the model run

python train.py

Optional Arguments are:

--learning_rate          The learning rate for MAML
--epsilon                The value of epsilon for updating model parameters in Reptile
--epochs                 Number of epochs
--K_shot_learning        How many sentences to sample for learning i.e 1-shot, 5-shot etc
--N_way_learning         How many tasks to test on
--hidden_size            Hidden size for LSTM units
--training_mode          MAML, Reptile which one to use for training the model
--inner_gradient_update  Number of inner gradient updates to perform