Training the models

Before following this guide, please make sure you installed all dependencies as described in the setup guide for your platform. A list of setup guides can be found here.

The sentiment and pattern models are the easiest to train. For this, cd into the EVE repository and run the following commands. Note that python is expected to be Python 3.6 or newer. If Python 3 is installed as python3 on your platform, run these commands with python3 instead.

python -m bot train-patterns
python -m bot train-sentiments

The generative chat model requires a chat corpus to train on. A chat corpus compatible with this project can be generated from German iOS or Android WhatsApp chat exports by using our parser. See the parser's README for information on how to generate the chat corpus from your chat exports. Next, copy the chat-data folder generated by the parser to eve/bot/, where eve/ is the path of the EVE repository. Now we can cd into the repository and run the following command to start training the chat model:

python -m bot train-chat

Again, python has to be changed to python3 if that's the name of Python 3.6 or newer on your platform. Note that the training of the generative chat model can take several hours. With our test hardware, training took two hours when using an Nvidia GeForce GTX 970 (GPU) or eight hours when using an Intel i5 4560 @ 3.5GHz (CPU). Alternatively, you can look into using Google Colab for training. An example of this can be seen in this tutorial.