def test_speedup_transform64_op_version(self):
data = load_iris()
X, _ = data.data, data.target
spd = OnnxSpeedupTransformer(PCA(),
target_opset=self.opset(),
enforce_float32=False)
spd.fit(X)
opset = spd.op_version
self.assertGreater(self.opset(), opset[''])
def test_speedup_transform64_onnx(self):
data = load_iris()
X, _ = data.data, data.target
spd = OnnxSpeedupTransformer(PCA(),
target_opset=self.opset(),
enforce_float32=False)
spd.fit(X)
expected = spd.transform(X)
onx = to_onnx(spd, X[:1])
oinf = OnnxInference(onx)
got = oinf.run({'X': X})['variable']
self.assertEqualArray(expected, got)
def test_speedup_transform64(self):
data = load_iris()
X, _ = data.data, data.target
spd = OnnxSpeedupTransformer(PCA(),
target_opset=self.opset(),
enforce_float32=False)
spd.fit(X)
spd.assert_almost_equal(X)
def test_speedup_transform32_numpy(self):
data = load_iris()
X, _ = data.data, data.target
spd = OnnxSpeedupTransformer(PCA(),
target_opset=self.opset(),
runtime="numpy")
spd.fit(X)
spd.assert_almost_equal(X, decimal=5)
def test_speedup_transform32_weight(self):
data = load_iris()
X, y = data.data, data.target
spd = OnnxSpeedupTransformer(StandardScaler(),
target_opset=self.opset())
w = numpy.ones(y.shape, dtype=X.dtype)
spd.fit(X, sample_weight=w)
spd.assert_almost_equal(X, decimal=5)
def test_speedup_transform32_numba(self):
data = load_iris()
X, _ = data.data, data.target
X = X.astype(numpy.float32)
spd = OnnxSpeedupTransformer(PCA(),
target_opset=self.opset(),
runtime="numba")
spd.fit(X)
spd.assert_almost_equal(X, decimal=5)
self.assertIn("CPUDispatch", str(spd.onnxrt_.func))
def test_speedup_transform64_pickle(self):
data = load_iris()
X, _ = data.data, data.target
spd = OnnxSpeedupTransformer(PCA(),
target_opset=self.opset(),
enforce_float32=False)
spd.fit(X)
st = BytesIO()
pickle.dump(spd, st)
st2 = BytesIO(st.getvalue())
spd2 = pickle.load(st2)
expected = spd.transform(X)
got = spd2.transform(X)
self.assertEqualArray(expected, got)
expected = spd.raw_transform(X)
got = spd2.raw_transform(X)
self.assertEqualArray(expected, got)
import matplotlib.pyplot as plt
from sklearn.datasets import make_regression
from sklearn.decomposition import PCA
from pyquickhelper.pycode.profiling import profile
from mlprodict.sklapi import OnnxSpeedupTransformer
from cpyquickhelper.numbers.speed_measure import measure_time
from tqdm import tqdm
################################
# Data and models to test.
data, _ = make_regression(1000, n_features=20)
data = data.astype(numpy.float32)
models = [
('sklearn', PCA(n_components=10)),
('python', OnnxSpeedupTransformer(PCA(n_components=10), runtime='python')),
('onnxruntime1',
OnnxSpeedupTransformer(PCA(n_components=10), runtime='onnxruntime1')),
('numpy', OnnxSpeedupTransformer(PCA(n_components=10), runtime='numpy')),
('numba', OnnxSpeedupTransformer(PCA(n_components=10), runtime='numba'))
]
#################################
# Training.
for name, model in tqdm(models):
model.fit(data)
#################################
# Profiling of runtime `onnxruntime1`.