Towards Safe Automated Refactoring of Imperative Deep Learning Programs to Graph Execution

TL;DR

This paper introduces an automated refactoring approach that helps developers convert imperative deep learning code to graph execution safely and efficiently, improving performance while preserving semantics.

Contribution

It presents a novel tensor analysis for imperative DL code and a refactoring method to determine when and how to safely migrate code to graph execution.

Findings

Refactoring preconditions can identify safe migration points.

The approach uses a tensor analysis tailored for imperative DL code.

Implementation as a PyDev plugin demonstrates practical applicability.

Abstract

Efficiency is essential to support responsiveness w.r.t. ever-growing datasets, especially for Deep Learning (DL) systems. DL frameworks have traditionally embraced deferred execution-style DL code -- supporting symbolic, graph-based Deep Neural Network (DNN) computation. While scalable, such development tends to produce code that is error-prone, non-intuitive, and difficult to debug. Consequently, more natural, less error-prone imperative DL frameworks encouraging eager execution have emerged at the expense of run-time performance. Though hybrid approaches aim for the "best of both worlds," using them effectively requires subtle considerations to make code amenable to safe, accurate, and efficient graph execution -- avoiding performance bottlenecks and semantically inequivalent results. We present our ongoing work on an automated refactoring approach that assists developers in specifying whether and how their otherwise eagerly-executed imperative DL code could be reliably and efficiently executed as graphs at run-time in a semantics-preserving fashion. The approach, based on a novel tensor analysis specifically for imperative DL code, consists of refactoring preconditions for automatically determining when it is safe and potentially advantageous to migrate imperative DL code to graph execution and modifying decorator parameters or eagerly executing code already running as graphs. The approach is being implemented as a PyDev Eclipse IDE plug-in and uses the WALA Ariadne analysis framework. We discuss our ongoing work towards optimizing imperative DL code to its full potential.