어제는 RAPIDS에 대한 소개와 RAPIDS APIs 중 cuDF 에 대한 예제에 대한 포스팅을 했습니다.
오늘은 RAPIDS APIS 중 cuML 에 대한 예제 포스팅을 간.단.하.게 해보겠습니다.
비교를 위해 KNN Classifier를 준비하였고 성능 비교를 위해 Scikit-learn 을 사용하였습니다.
# %%
# RAPIDS cuML kNN model
import time
import cudf, cuml
import pandas as pd
from cuml.neighbors import KNeighborsClassifier as cuKNeighbors
from sklearn.neighbors import KNeighborsClassifier as skKNeighbors
print("========================================")
print("========================================")
print("============= Using cuML ===============")
print("========================================")
print("========================================")
train = cudf.read_csv('./mnist/train.csv')
test = cudf.read_csv('./mnist/test.csv')
start = time.time()
model = cuKNeighbors(n_neighbors=7)
model.fit(train.iloc[:,1:785], train.iloc[:,0])
# y_hat = model.predict(test) # Exception occured (version 0.14.0)
print(f"Elapsed Time: {time.time()-start}\n")
print("========================================")
print("========================================")
print("========== Using Scikit-learn ==========")
print("========================================")
print("========================================")
train = pd.read_csv('./mnist/train.csv')
test = pd.read_csv('./mnist/test.csv')
start = time.time()
model = skKNeighbors(n_neighbors=7)
model.fit(train.iloc[:,1:785], train.iloc[:,0])
# y_hat = model.predict(test)
print(f"Elapsed Time: {time.time()-start}")
========================================
========================================
============= Using cuML ===============
========================================
========================================
Elapsed Time: 0.3921339511871338
========================================
========================================
========== Using Scikit-learn ==========
========================================
========================================
Elapsed Time: 23.88076663017273
Test 시 성능도 비교하려했으나 cuML 버전 에러로 인해 테스트는 못했습니다.
그래도 fit 부분에서 매우 큰 차이를 보입니다!
역시나….GPU네요!
