step 1
conda set up -n base -c conda-forge mamba
step 2
mamba create -n rapids -c rapidsai -c nvidia -c conda-forge python=3.10 cudatoolkit=11.8 cuml=24.02
notification: alone machine, cuda 12.0 all the time throw error, I don’t know why, though my nvidia-smi exhibits cuda 12.4
step 3
conda activate rapids
pip set up ipykernel
python -m ipykernel set up --name rapids --user
mamba set up -c conda-forge scikit-learn
Benchmark
import cuml
import pandas as pdfrom cuml.benchmark.runners import SpeedupComparisonRunner
from cuml.benchmark.algorithms import algorithm_by_name
import warnings
warnings.filterwarnings('ignore', 'Anticipated column ')
print(cuml.__version__)
N_REPS = 3 # Variety of instances every check is repeated
DATA_NEIGHBORHOODS = "blobs"
DATA_CLASSIFICATION = "classification"
DATA_REGRESSION = "regression"
INPUT_TYPE = "numpy"
benchmark_results = []
SMALL_ROW_SIZES = [2**x for x in range(14, 17)]
LARGE_ROW_SIZES = [2**x for x in range(18, 24, 2)]
SKINNY_FEATURES = [32, 256]
WIDE_FEATURES = [1000, 10000]
VERBOSE=True
RUN_CPU=True
def enrich_result(algorithm, runner, consequence):
consequence["algo"] = algorithm
consequence["dataset_name"] = runner.dataset_name
consequence["input_type"] = runner.input_type
return consequence
def execute_benchmark(algorithm, runner, verbose=VERBOSE, run_cpu=RUN_CPU, **kwargs):
outcomes = runner.run(algorithm_by_name(algorithm), verbose=verbose, run_cpu=run_cpu, **kwargs)
outcomes = [enrich_result(algorithm, runner, result) for result in results]
benchmark_results.lengthen(outcomes)
runner = cuml.benchmark.runners.SpeedupComparisonRunner(
bench_rows=SMALL_ROW_SIZES,
bench_dims=SKINNY_FEATURES,
dataset_name=DATA_NEIGHBORHOODS,
input_type=INPUT_TYPE,
n_reps=N_REPS,
)execute_benchmark("NearestNeighbors", runner)
