Semi-supervised learning research - Fall 2020 and Spring 2021

The preliminary results were presented at LMU's Undergraduate Research Symposium 2021 and This is Honors, under the title Mapping of exercise logs to a database using Neural Networks and data augmentation techniques. The full presentation can be found here.

Install dependecies

For running the models

Install pytorch and allenlp.

For hyperparameter tunning install optuna and optuna dashboard.

For data augmentation

Install NLP AUG, and related libraries (They can be found at the same github repo).

Train models with Allen NLP

How to convert jsonnet files into json

jsonnet [cofig_file] -o [output_file]

How to run experiments

allennlp train [config_file] -s [results-dir]

How to use the models

allennlp predict [model_dir]/model.tar.gz [data_file_path].json --predictor [predictor_registered_name]

Visualize the results with TensorBoard

How to show results locally

tensorboard --logdir=[log_dir]

How to upload results

tensorboard dev upload --logdir=[log_dir]

Perform hyperparameter optimization with Optuna

How to run the optimization

Run file containing optuna code: python hyperparam_optim.py

How to vizualize the optimization

optuna-dashboard sqlite:[path-to-.db-file]

Repo Structure

experiments

All the data, models and results are within this directory. For more information read the particular README for this directory.