ChatLearner

bshao001/ChatLearner is ranked 17,380 among these 8 peers; l0ng-ai/papr is ranked 17,387.

ChatLearner A chatbot implemented in TensorFlow based on the new sequence to sequence (NMT) model, with certain rules seamlessly integrated. **For those who are interested in chatbots in Chinese, please check here. ** The core of ChatLearner (Papaya) was built on the NMT model( which has been adapted here to fit the needs of a chatbot. 4 and many other changes since TensorFlow 1. 12, this ChatLearner version only supports TF version 1. Easy updates can be made in tokenizeddata. py

file if you need to support TensorFlow 1. Before starting everything else, you may want to get a feeling of how ChatLearner behaves. Take a look at the sample conversation below or here, or if you prefer to try my trained model, download it here. rar file, and copy the Result folder into the Data folder under your project root. txt file is also included in case I update it without updating the trained model in the future. Highlights and

Specialties: Why do you want to spend time checking this repository? The Papaya Data Set for training the chatbot. You can easily find tons of training data online, but you cannot find any with such high quality. See the detailed description below about the data set. The concise code style and clear implementation of the new seq2seq model based on dynamic RNN (a. It is customized for chatbots and much easier to understand compared with the official tutorial. The idea

of using seamlessly integrated ChatSession to handle basic conversational context. Some rules are integrated to demo how to combine traditional rule-based chatbots with new deep learning models. No matter how powerful a deep learning model can be, it cannot even answer questions requiring simple arithmetic calculations, and many others. The approach demonstrated here can be easily adapted to retrieve news or other online information. With the rules implemented, it can then properly answer many interesting questions. " It can then

present stories and jokes randomly and not being limited by the sequence length of the decoder. "How much is twelve thousand three hundred four plus two hundred fifty six? " or "How much is twelve thousand three-hundred and four divided by two-hundred-fifty-six? " or "If x=55 and y=19, how much is y - x? " or "How much do you get if you subtract eight from one hundred? " or even "If x = 99 and y = 228 / x, how

much is y? " If you are not interested in rules, you can easily remove those lines related to knowledgebase. A SOAP-based web service (and a REST-API-based alternative, if you don't like to use SOAP) allows you to present the GUI in Java, while the model is trained and running in Python and TensorFlow. A simple solution (in-graph) to convert a string tensor to lower case in TensorFlow. It is required if you utilize the new DataSet

API (tf. TextLineDataSet) in TensorFlow to load training data from text files. The repository also contains a chatbot implementation based