rpa-models

This repository ONLY contains the datasets and scripts used to train the models in RPA Tomorrow. Currently only two categories of models are under development:

• deep (using seq2seq)
• simple (smaller models trained from scratch)
  • simple-email
  • simple-calendar
  • simple-reminder

Requirements

• Python 3.8.5
• Anaconda

Setup

Load the Conda environment.

 conda activate rpa-models

You will need pre-trained Spacy models for the Deep model.

 python -m spacy download en_core_web_lg

Training models

To train the deep model with the default configurations run

  python deep/seq2seq.py #or

To train the simple model, run the following with the dataset you want to train the model only

  python simple/train_model.py -d <DATASET_NAME> -o <OUTPUT_DIR>

The DATASET_NAME is the name of the directory you want to work on inside simple/datasets e.g. "reminder". If you don't specify the output directory the model will not be stored.

NOTE!

For the pre-trained models and their releases, see model-releases