Knnbot: a KNN Messenger Bot

Knnbot is a Facebook Messenger bot written in Python, capable of training a machine learning algorithm known as K-Nearest Neighbor using manually inputted data from a user. This project is more on the experimental side, a simple proof of concept where I wanted to show a use case of how a machine learning algorithm could be applied to a chatbot.

Dependencies

• Flask==0.11.1
• requests==2.10.0
• numpy==1.11.1
• pymessenger==0.0.7.0 (PyMessenger)
• pillow==3.1.1
• matplotlib==1.5.3

With pip run pip install -r requirements.txt

Regarding the algorithm

For simplicity purposes the bot just accepts feature vectors of two dimensions whose features are integers. This could be easily changed by just modifying the input validation in the function add_to_training

The algorithm itself has no limitations regarding this.

Features

The bot accepts the command status which will print the number of training examples the user has inputted, the K, and the state in which the bot is currently at, which is either training if the user is training the system, or predict is the user has enough training examples.

Also, there will be two buttons: training classes, and show knn. The first one will show all the different classes seen during training and the number of cases associated with that label. The second button will display a 2D scatterplot of the training examples which each data point colored depending on its class.

Status button

Different classes and their frequencies

Plot of how the KNN model looks

Instructions

At start, Knnbot will in the training state. During this state the user have to input a feature vector made of two integers, and the class or label, which should also be an integer. For example 10,5,2 is a feature vector [10,5] with class 2.

Once the user has 3 or more training examples, Knnbot will ask you if you which to test the system. If the answer is yes, then Knnbot will be in predict state. While in this state, Knnbot will predict the class of your input which should be of the form x,y where x and y are integers; this input is the feature vector.

To switch back to the training or predict state, type the command train or predict.

To recap, these are all the commands. During TRAINING state:

• To create a new training example, create a new feature vector of the shape x,y,z where x and y are the features, and z is the class. During TEST state:
• To test the system, write a feature vector of the shape x,z where x and y are the features.

During any state:

• status to see several information regarding the training
• train to enter the training state
• test to enter the testing state (if the bot has more than K training examples).