GraphGrail Ai – is the world’s first Artificial Intelligence platform for Blockchain built on top of Natural Language Understanding technology with the DApps marketplace.

Retrieval-Based Conversational Model in Tensorflow (Ubuntu Dialog Corpus)

Please read the blog post for this code

Overview

The code here implements the Dual LSTM Encoder model from The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems.

Setup

This code uses Python 3. Clone the repository and install all required packages:

1. install tensorflow (version 0.11 and above wokr correctly, version 0.10 not tested)
2. (optional) install cuda + cudnn (for gpu support)
2. pip3 install -U pip
3. pip3 install -r requirements.txt

Get the Data

Download the train/dev/test data here and extract the acrhive into ./data.

Training

python3 udc_train.py
or------------------
sh train.sh

Evaluation

python3 udc_test.py --model_dir=...
or------------------
sh test.sh

example:

python3 udc_test.py --model_dir=./runs/1481183770/
or------------------
sh predict.sh

Evaluation

python3 udc_predict.py --model_dir=...

example:

python3 udc_predict.py --model_dir=./runs/1481183770/

Issues

• if you have problem's with loading CUDA library libcuda.so.1 use *.sh script, or export variables in bash:

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"
export CUDA_HOME=/usr/local/cuda

• if you have multiple gpu devices and expecting troubles with performance, manualy select device in bash:

export CUDA_VISIBLE_DEVICES=0

• if you have error (see below), you maybe use trained model from other machine, and you must retrain model on own machine

tensorflow.python.framework.errors.NotFoundError: /home/user/git/chatbot/chatbot-retrieval/runs/1481104318

This module is not belong to Graph Grail!!! It will be used to integrate with the micro services provided by Graph Grail.