/meta-learned-embeddings

Meta-Reinforcement Learning For Learning Domain Specific Embeddings While Utilizing Cross-Domain Knowledge

Primary LanguageJupyter Notebook

meta-learned-embeddings

Meta-Reinforcement Learning For Learning Domain Specific Embeddings While Utilizing Cross-Domain Knowledge

NECESSARY INSTALLS:

  • Create conda environment
  • pip install tensorboardx
  • pip install pytorch-pretrained-bert

HOW TO RUN MAML W/ BERT INITIALIZATION:

  • cd to meta-learned-embeddings directory
  • python bert_main.py
  • Results will be in maml_output/few_shot directory
  • To run tensorboard: tensorboard --logdir maml_output/few_shot/

HOW TO RUN BERT W/O MAML:

  • cd to meta-learned-embeddings directory
  • python bert_data_preprocessing.py
  • Zero-Shot:
    • python pytorch-pretrained-BERT/examples/run_classifier_zero_shot.py --do_train --do_eval
    • To run tensorboard: tensorboard --logdir bert_zero_shot_output/tb
  • Few-Shot:
    • python pytorch-pretrained-BERT/examples/run_classifier_few_shot.py --do_train --do_eval
    • To run tensorboard: tensorboard --logdir bert_few_shot_output/tb
  • Standard:
    • python pytorch-pretrained-BERT/examples/run_classifier_standard.py --do_train --do_eval
    • To run tensorboard: tensorboard --logdir bert_standard_output/tb
  • Results will be in bert_zero_shot_output, bert_few_shot_output, and bert_standard_output folders