PypagAI is a framework that aims to quickly develop new QA models, test new datasets and reproduce experiments. The main objective of the framework is to speed up the development of new QA models. Thus, it is necessary to normalize experiments, prepare data, and create a repository of QA models. By building and publishing this framework, we would like to lower the entry barrier for more people to try out.
PypagAI was inspired on Facebook's dialog framework, ParlAI, but it is easier and faster to test models. It uses Sacred as experiments backend and is very easy to integrate new Keras and TensorFlow models
To run with default configurations
python -m experiment.qa_experiment -uIf you want list all available parameters just type
python -m experiment.qa_experiment print_configChanging the models
python -m experiment.qa_experiment with model_default_cfg.model=pypagai.models.model_rnn.RNNModel -u
Changing dataset
python -m experiment.qa_experiment with dataset_cfg.task=3 -uYou can also override the default experiment main and create your own flow. Just load data and model libraries.