def kernel(request):
if request.param == 1:
return GaussianKernel(sigma=1)
elif request.param == 2:
return LinearKernel()
elif request.param == 3:
return PolynomialKernel(1.2, 3, 2.5)
def k1():
return GaussianKernel(sigma=1)
def kernel(self):
return GaussianKernel(100.0)
globals=_vars,
number=1,
repeat=exp['repetitions'])
exp['timings'].append(min(timings))
print(exp, flush=True)
torch.cuda.empty_cache()
return experiments
if __name__ == "__main__":
aparse = argparse.ArgumentParser(description="MMV experiment runner")
aparse.add_argument('--num-gpus', type=int, required=True)
args = aparse.parse_args()
num_gpus = args.num_gpus
kernel = GaussianKernel(3.0)
Ns = [
1000, 5000, 20000, 50000, 100000, 200000, 400000, 600_000, 1_000_000,
2_000_000, 10_000_000, 50_000_000, 100_000_000
]
KeopsDs = [10, 50, 100, 250, 500, 750, 1000, 1250]
OurDs = [
10, 50, 100, 250, 500, 750, 1000, 1250, 1500, 2000, 2500, 3000, 4000,
5000, 7000, 10000
]
defaultM = 20_000
defaultN = 20_000
defaultT = 10
defaultD = 10
experiments = [
def kernel():
return GaussianKernel(10.0)