simic_pool

pool process for simic extension

This program is a server process that communicates with the jupyterlab extension Simic via TCP.

Run this program separately from jupyter lab

Before running, please follow the below steps.

1. run the following shell file to clone the repositories.

zsh clone.sh

or

bash clone.sh depending on your shell

2. download the pool.tree and d2v.model files from the following links.

https://drive.google.com/file/d/1hx7QlTOi93N1_a6Jo6wb5LaeP3f6kGQC/view?usp=sharing

place this file at the project root

https://drive.google.com/file/d/1I_1hhyF5jnN7RQdBzemB0BOw_ywK2Hpq/view?usp=share_link

place this file at simic_pool/train/

3. run

make build

for gradle module.

Then finally run

python3 server.py

IP and Port statically set for now

Implementation

Simic in jupyterlab side is composed of FrontEnd and Server Extension (Backend).

Simic Pool is a separate application that TCP communicates with the Simic's Server Extension.

Simic Pool is designed to perform 2 main tasks.

1. Detreefy

Change Vector collected from 20 ML/DL projects are pickled as a tree data structure.

Meaning that if a vector is 102, each node represents one scalar, namely 1, 0, and 2 as below.

1 -- 0 -- 2
   |
    -1 -- 3
   |
    -2 -- 4

Simic Pool uses level order traversal for finding an exact match

If it fails to find the exact match, it looks for the closest child node -> sibling node -> parent node, depending on the vector size (not implemented yet)

2. Doc2Vec Cosine Similarity

If the found node contains multiple changes, the Simic Pool tokenizes the user input and the data found from the detreefication to compute similarity on a model trained with over 20000 source codes.