PlugML

A PennApps XXII submission by the Carnegie Felons

Story

Inspiration

Once upon a time, a systems engineer wanted to learn machine learning. They tried and failed. Then, they met a machine learning engineer, and together they brought together a team to make machine learning easy.

What it does

PlugML provides two components: the PlugML Engine and the Endpoint Library. The engine is a hands-on visual application that allows users to easily design neural networks. App developers can then use the very simple endpoint library to connect their applications through a pipeline to the engine in real time.

The engine has two modes: training and testing. In training mode, the engine will consume data points and their corresponding labels from the pipeline to develop a learnt neural network model. In testing mode, the engine will be able to make predictions on input data from the pipeline based on the model and transmit the results back through the pipeline to the app.

How we built it

The Endpoint Library is a simple python library compressed into a whl file for easy pip install.

The PlugML Engine is a tkinter application with visuals displayed on canvas. The machine learning back-end is powered by tensorflow. The pipeline between the engine and a developed app is implemented using TCP sockets.

Challenges we ran into

UI.

Accomplishments that we're proud of

Successfully integrating a neural network into a simple, easy-to-use user interface.

What we learned

UI is hard and requires systematic planning and delegation.

What's next for PlugML

More ML features! (CNN, DQN, KNN, A*, etc.)

Usage

Git clone the repo and run python src/main.py to launch the engine. Here is the code for the sample projects used in the presentation video:

Sine curve (-4, 4)

Training

import math
import plugml.ipc as ipc
import random

x_data = []
y_data = []
for i in range(int(input('train size: '))):
  x = random.uniform(-4, 4)
  y = math.sin(x)
  x_data.append(x)
  y_data.append(y)

ipc.connect(int(input('port: ')))
ipc.transmit_list(x_data)
ipc.transmit_list(y_data)

Testing

import math
import plugml.ipc as ipc
import random

port = int(input('port: '))
while True:
  ipc.connect(port)
  x = random.uniform(-4, 4)
  y = math.sin(x)
  ipc.transmit_list([x])
  data = ipc.retrieve_mapped_list(float)
  prediction = data[0]
  error = (prediction - y)**2
  print('x:', x)
  print('y', y)
  print('predict', prediction)
  print('err', error)
  print('')