Autoslim

Autoslim is a framework for AI graph pruning. It generates synthetic state-transition graphs, converts them into structured CSVs, prunes edges based on score thresholds and machine learning, and exports results for analysis or hardware simulation.

Features

• Generate random XML graphs with configurable size and fan-out.
• Convert XML → CSV with state transitions and scores.
• Prune graphs using thresholds and Random Forest models.
• Export results to plots, logs, and Memory Initialization Files (MIF).

Makefile Overview

The Makefile automates the full workflow: make → compiles the C++ tools (gen_xml_multiple_files4, generate_csv_multiple). make genFolder → creates experiment folders for different dataset sizes (1K–64K). make run → generates graphs, converts to CSV, and runs pruning for all dataset sizes. make beforemif → generate MIF files from raw CSVs. make prunedmif → generate MIF files from pruned CSVs. make clean → remove generated binaries and results.

Usage

Typical workflow:

Compile tools

make

Create experiment folders

make genFolder

Run full pipeline (generate, convert, prune) for 1K–64K graphs

make run

Export MIF files (before or after pruning)

make beforemif
make prunedmif

Results are organized in folders like 1A_xmlFiles, 1A_csvFiles, 1A_prunedResults, etc.