graphsense 0.0.1

graphsense is a library to train and infer graph embedding based code completion models

Requirements

• Python 3.8 or greater

installation:

pip install graphsense

Training example:

from graphsense import GraphSense

g = GraphSense()

g.line_completion(input_path="code_files", output_path="output")

Inference example:

from graphsense import GraphSense

g = GraphSense()

g.load_model("output/graph_embeddings.model")
next = g.infer("def factorial(n):")

print("next item predicted: ", next)

Performance Comparison with gpt2_medium finetuned model

Dataset used to train models: https://github.com/TheAlgorithms/Python

gpt2_medium finetuned model

input: def factorial(n)  
output: return 1 if n == 1 else n * factorial(n - 1)   
model size: 1.44GB   
avg inference time: 10.3302 seconds  
CPU Usage: 8.3%  
Memory Usage: 68.54 MB  

graph embedding model

input: def factorial(n)  
output: return 1 if n == 1 else n * factorial(n - 1)   
model size: 13.2MB
avg inference time: 2.1870 seconds  
CPU Usage: 0.2%
Memory Usage: 4.54 MB