def test_onnx_stats(self):
iris = load_iris()
X, y = iris.data, iris.target
X_train, X_test, y_train, _ = train_test_split(X, y, random_state=11)
clr = LogisticRegression()
clr.fit(X_train, y_train)
temp = get_temp_folder(__file__, "temp_onnx_stats")
data = os.path.join(temp, "data.csv")
pandas.DataFrame(X_test).to_csv(data, index=False)
pkl = os.path.join(temp, "model.pkl")
with open(pkl, "wb") as f:
pickle.dump(clr, f)
outonnx = os.path.join(temp, 'outolr.onnx')
convert_validate(pkl=pkl,
data=data,
verbose=0,
method="predict,predict_proba",
outonnx=outonnx,
name="output_label,output_probability")
st = BufferedPrint()
main(args=["onnx_stats", "--name", outonnx], fLOG=st.fprint)
res = str(st)
self.assertIn("ninits: 0", res)
def test_convert_validate(self):
iris = load_iris()
X, y = iris.data, iris.target
X_train, X_test, y_train, _ = train_test_split(X, y, random_state=11)
clr = LogisticRegression()
clr.fit(X_train, y_train)
temp = get_temp_folder(__file__, "temp_convert_validate")
data = os.path.join(temp, "data.csv")
pandas.DataFrame(X_test).to_csv(data, index=False)
pkl = os.path.join(temp, "model.pkl")
with open(pkl, "wb") as f:
pickle.dump(clr, f)
res = convert_validate(pkl=pkl,
data=data,
verbose=0,
method="predict,predict_proba",
name="output_label,output_probability")
self.assertIsInstance(res, dict)
self.assertLess(res['metrics'][0], 1e-5)
self.assertLess(res['metrics'][1], 1e-5)
self.assertRaise(
lambda: convert_validate(pkl=pkl,
data=data,
verbose=0,
method="predict,predict_proba",
name="output_label"), ValueError)
self.assertRaise(
lambda: convert_validate(pkl=pkl,
data=data,
verbose=0,
method="predict,predict_probas",
name="output_label,output_probability"),
AttributeError)
self.assertRaise(
lambda: convert_validate(pkl=pkl,
data=data,
verbose=0,
method="predict,predict_proba",
name="output_label,output_probabilities"),
KeyError)
res = convert_validate(pkl=pkl,
data=data,
verbose=0,
method="predict,predict_proba",
name="output_label,output_probability",
noshape=True)
self.assertIsInstance(res, dict)
self.assertLess(res['metrics'][0], 1e-5)
self.assertLess(res['metrics'][1], 1e-5)
def test_cli_convert_validater_switch(self):
iris = load_iris()
X, y = iris.data, iris.target
X_train, X_test, y_train, _ = train_test_split(X, y, random_state=11)
clr = LogisticRegression()
clr.fit(X_train, y_train)
temp = get_temp_folder(__file__, "temp_cli_convert_validate_switch")
data = os.path.join(temp, "data.csv")
pandas.DataFrame(X_test).to_csv(data, index=False)
pkl = os.path.join(temp, "model.pkl")
with open(pkl, "wb") as f:
pickle.dump(clr, f)
res = convert_validate(pkl=pkl, data=data, verbose=0,
method="predict,predict_proba",
name="output_label,output_probability")
st = BufferedPrint()
args = ["convert_validate", "--pkl", pkl, '--data', data,
'--method', "predict,predict_proba",
'--name', "output_label,output_probability",
'--verbose', '1', '--use_double', 'switch']
main(args, fLOG=st.fprint)
res = str(st)
self.assertIn(
"[convert_validate] compute predictions with method 'predict_proba'", res)
def test_cli_convert_validater_pkl_nodata(self):
temp = get_temp_folder(__file__,
"temp_cli_convert_validate_pkl_nodata")
monx = os.path.join(temp, "gpr.onnx")
pkl = os.path.join(temp, "booster.pickle")
if not os.path.exists(pkl):
return
st = BufferedPrint()
res = convert_validate(pkl=pkl,
data=None,
verbose=0,
method="predict,predict_proba",
name="output_label,output_probability",
outonnx=monx,
fLOG=st.fprint)
res = str(st)
self.assertNotIn("[convert_validate] compute predictions", res)
def test_cli_convert_validater_float64_gpr(self):
iris = load_iris()
X, y = iris.data, iris.target
X_train, X_test, y_train, _ = train_test_split(X, y, random_state=11)
clr = GaussianProcessRegressor()
clr.fit(X_train, y_train)
temp = get_temp_folder(__file__,
"temp_cli_convert_validate_float64_gpr")
monx = os.path.join(temp, "gpr.onnx")
data = os.path.join(temp, "data.csv")
pandas.DataFrame(X_test).to_csv(data, index=False)
pkl = os.path.join(temp, "model.pkl")
with open(pkl, "wb") as f:
pickle.dump(clr, f)
try:
res = convert_validate(
pkl=pkl,
data=data,
verbose=0,
method="predict",
name="GPmean",
use_double='float64',
options="{GaussianProcessRegressor:{'optim':'cdist'}}")
except RuntimeError as e:
if "requested version 10 < 11 schema version" in str(e):
return
raise e
self.assertNotEmpty(res)
st = BufferedPrint()
args = [
"convert_validate", "--pkl", pkl, '--data', data, '--method',
"predict", '--name', "GPmean", '--verbose', '1', '--use_double',
'float64', '--options',
"{GaussianProcessRegressor:{'optim':'cdist'}}", '--outonnx', monx
]
main(args, fLOG=st.fprint)
res = str(st)
self.assertExists(monx)
with open(monx, 'rb') as f:
model = onnx.load(f)
self.assertIn('CDist', str(model))
