A demo deep learning RNN chatbot using Tensorflow with a C# WPF GUI.

Note: Pretrained models are not provided. Please train your own network.

To use the python backend, run socket.py and connect to the socket on localhost:8089. Then, send the user input. For example, Hello World!. The socket should return the result as a simple string (no parsing needed).

First, start the python socket server with the directions above. Then, start the GUI and connect to the socket with the ip localhost and port 8089. Once the socket connects, use the first tab to talk with the chatbot.

• Tensorflow
• Numpy
• socket (should be installed with python)
• re (should be installed with python)

To install Tensorflow, run

pip install tensorflow

For the gpu version of Tensorflow, run

pip install tenorflow-gpu

To install Numpy, run

pip install numpy

Training a chatbot network will take a roughly a few hours. In order to train, you must provide a word embedding (the scripts by default use glove.twitter.27B.100d.txt) and the Cornell movie dialog corpus txts in the ./Data folder.

To train the network, run Train.py and input the number of epochs to train for. Input save to save the network for the socket server and exit to exit (without saving). Input predict for inference.

Train epochs: 10  # train 10 epochs
Train epochs: predict hey! how are you  # inference with input sequence 'hey! how are you'
Train epochs: save  # saves the network at ./model
Train epochs: 15
Train epochs: save
Train epochs: exit

Apache License 2.0