Chatbot using Keras

Here we are building a chatbot and this chatbot can have resoning capability.

Dataset Information

Here we are using bAbI dataset which is buit by Facebook AI Research.

Dependencies

• Python 2.7
• TensorFlow 1.4.1: Refer this link
• keras: Refer this link
• functools: It's python standard library.
• tarfile: It's python standard library.
• re : It's python standard library.
• h5py: $ sudo pip install h5py

Usage

• The dataset used here is babi-tasks-v1-2. Link of the dataset is here, its a relatively small dataset but a great dataset nonetheless

• In main.py file there are following parameters which can be change in following manner to train and test the model

• We are using the concepts of memory network and it is LSTM based models performed better than GRU based models for this task.

For Training

Step 1: Open main.py

Step 2: For training, set the parameters as given below.
train_model = 1         #(1 means training mode and 0 means no training mode) 
train_epochs = 100
load_model = 0          #( 1 means load the trained model and 0 means doesn't load trained model)
batch_size = 32
lstm_size = 64
test_qualitative = 0    #(1 means test trained on randomly generated story and 0 means do not perform test on ramdomly generated story)
user_questions = 0      #(1 means test trained on randomly generated story and 0 means do not perform test on ramdomly generated story)

Step 3: Run main.py

For Testing

Here we can perform two types of testing.

• Testing for randomly generated story
• Testing for used given story

Testing for randomly generated story

Step 1: Open main.py

Setp 2: For testing ramdomly generated story, set the parameters as given below.

train_model = 0
train_epochs = 100
load_model = 1
batch_size = 32
lstm_size = 64
test_qualitative = 1
user_questions = 0

Step 3: Run main.py

Testing for used given story

Step 1: Open main.py

Setp 2: For testing user given story, set the parameters as given below.

train_model = 0
train_epochs = 100
load_model = 1
batch_size = 32
lstm_size = 64
test_qualitative = 0
user_questions = 1

Step 3: Run main.py

Credit

Credit for the majority of code here goes to Batchu Vishal. I've merely created a wrapper to get people started.