| """Evaluator for matrix multiplication tensor decomposition (n=2, m=4, p=5).""" |
| import subprocess, sys, json |
|
|
| BENCHMARK = 32 |
|
|
| def evaluate(program_path): |
| script = f""" |
| import json, sys, os, time |
| import numpy as np |
| sys.path.insert(0, os.path.dirname(r'{program_path}')) |
| module_name = os.path.splitext(os.path.basename(r'{program_path}'))[0] |
| start = time.time() |
| try: |
| program = __import__(module_name) |
| decomposition, n, m, p, loss, rank = program.run() |
| except Exception as e: |
| print(json.dumps({{"error": str(e)}})); sys.exit(0) |
| eval_time = time.time() - start |
| U, V, W = decomposition |
| U, V, W = np.array(U), np.array(V), np.array(W) |
| if U.shape != (n*m, rank) or V.shape != (m*p, rank) or W.shape != (n*p, rank): |
| print(json.dumps({{"error": f"bad shapes U={{U.shape}} V={{V.shape}} W={{W.shape}}"}})); sys.exit(0) |
| T = np.zeros((n*m, m*p, n*p)) |
| for i in range(n): |
| for j in range(m): |
| for k in range(p): |
| T[i*m+j, j*p+k, k*n+i] = 1 |
| R = np.einsum("ir,jr,kr->ijk", U, V, W) |
| if not np.array_equal(T, R): |
| print(json.dumps({{"error": "decomposition does not match tensor"}})); sys.exit(0) |
| score = {BENCHMARK} / rank |
| print(json.dumps({{"combined_score": score, "rank": rank, "eval_time": eval_time}})) |
| """ |
| try: |
| r = subprocess.run([sys.executable, "-c", script], capture_output=True, text=True, timeout=600) |
| for line in reversed(r.stdout.strip().splitlines()): |
| if line.strip().startswith("{"): return json.loads(line.strip()) |
| return {{"combined_score": 0.0, "error": r.stderr[-500:]}} |
| except Exception as e: |
| return {{"combined_score": 0.0, "error": str(e)}} |
|
|