def test_onnx_regressor():
importorskip('skl2onnx')
importorskip('onnxruntime')
from aict_tools.io import save_model
from aict_tools.onnx import ONNXModel
with tempfile.TemporaryDirectory(prefix='aict_tools_test_') as tmpdir:
model_path = os.path.join(tmpdir, 'model.onnx')
save_model(clf, feature_names, model_path, label_text='classifier')
model = ONNXModel(model_path)
assert model.meta['model_author'] == 'aict-tools'
assert model.meta['feature_names'].split(',') == feature_names
assert model.feature_names == feature_names
probas_onnx = model.predict_proba(X_clf)
probas_skl = clf.predict_proba(X_clf)
assert np.all(np.isclose(probas_onnx, probas_skl, rtol=1e-2,
atol=1e-4))
# make sure pickle is also saved
clf_load = joblib.load(model_path.replace('.onnx', '.pkl'))
assert clf_load.feature_names == feature_names
assert np.all(clf.predict(X_clf) == clf_load.predict(X_clf))
def test_onnx_regressor():
importorskip("skl2onnx")
importorskip("onnxruntime")
from aict_tools.io import save_model
from aict_tools.onnx import ONNXModel
with tempfile.TemporaryDirectory(prefix="aict_tools_test_") as tmpdir:
model_path = os.path.join(tmpdir, "model.onnx")
save_model(clf, feature_names, model_path, label_text="classifier")
model = ONNXModel(model_path)
assert model.meta["model_author"] == "aict-tools"
assert model.meta["feature_names"].split(",") == feature_names
assert model.feature_names == feature_names
probas_onnx = model.predict_proba(X_clf)
probas_skl = clf.predict_proba(X_clf)
assert np.all(np.isclose(probas_onnx, probas_skl, rtol=1e-2,
atol=1e-4))
# make sure pickle is also saved
clf_load = joblib.load(model_path.replace(".onnx", ".pkl"))
assert clf_load.feature_names == feature_names
assert np.all(clf.predict(X_clf) == clf_load.predict(X_clf))
def test_pmml_regressor():
importorskip('sklearn2pmml')
importorskip('jpmml_evaluator')
from aict_tools.pmml import PMMLModel
from aict_tools.io import save_model
with tempfile.TemporaryDirectory(prefix='aict_tools_test_') as tmpdir:
model_path = os.path.join(tmpdir, 'model.pmml')
save_model(reg, feature_names, model_path, label_text='regressor')
model = PMMLModel(model_path)
assert model.feature_names == feature_names
assert np.allclose(model.predict(X_reg), reg.predict(X_reg))
# make sure pickle is also saved
reg_load = joblib.load(model_path.replace('.pmml', '.pkl'))
assert reg_load.feature_names == feature_names
assert np.all(reg.predict(X_reg) == reg_load.predict(X_reg))
def test_pickle():
from aict_tools.io import save_model
with tempfile.TemporaryDirectory(prefix='aict_tools_test_') as tmpdir:
model_path = os.path.join(tmpdir, 'model.pkl')
save_model(clf, feature_names, model_path, label_text='classifier')
clf_load = joblib.load(model_path)
assert clf_load.feature_names == feature_names
with tempfile.TemporaryDirectory(prefix='aict_tools_test_') as tmpdir:
model_path = os.path.join(tmpdir, 'model.pkl')
save_model(reg, feature_names, model_path, label_text='regressor')
reg_load = joblib.load(model_path)
assert reg_load.feature_names == feature_names
assert np.all(reg.predict(X_reg) == reg_load.predict(X_reg))
assert np.all(clf.predict(X_clf) == clf_load.predict(X_clf))
def test_pmml_classifier():
importorskip('sklearn2pmml')
importorskip('jpmml_evaluator')
from aict_tools.pmml import PMMLModel
from aict_tools.io import save_model
with tempfile.TemporaryDirectory(prefix='aict_tools_test_') as tmpdir:
model_path = os.path.join(tmpdir, 'model.pmml')
save_model(clf, feature_names, model_path, label_text='classifier')
model = PMMLModel(model_path)
# order seems to be changing
assert model.feature_names == feature_names
proba = model.predict_proba(X_clf)
assert np.allclose(proba[:, 1], clf.predict_proba(X_clf)[:, 1])
# make sure pickle is also saved
clf_load = joblib.load(model_path.replace('.pmml', '.pkl'))
assert clf_load.feature_names == feature_names
assert np.all(clf.predict(X_clf) == clf_load.predict(X_clf))