def test_varimp():
train = h2o.upload_file(pyunit_utils.locate("smalldata/wine/winequality-redwhite-no-BOM.csv"))
y = "quality"
# get at most one column from each type
cols_to_test = []
for col, typ in train.types.items():
for ctt in cols_to_test:
if typ == train.types[ctt] or col == y:
break
else:
cols_to_test.append(col)
aml = H2OAutoML(seed=1234, max_models=5)
aml.train(y=y, training_frame=train)
assert aml.varimp(use_pandas=True).shape == (12, 5)
assert h2o.explanation.varimp(aml.leaderboard[aml.leaderboard["model_id"].grep("Stacked", invert=True, output_logical=True), :].head(3), num_of_features=3, use_pandas=True).shape == (3, 3)
varimp_1 = aml.varimp(use_pandas=False)
assert varimp_1[0].shape == (12, 5)
assert len(varimp_1[1]) == 5
assert len(varimp_1[2]) == 12
varimp_2 = h2o.explanation.varimp(aml.leaderboard[aml.leaderboard["model_id"].grep("Stacked", invert=True, output_logical=True), :].head(4), num_of_features=3, use_pandas=False)
assert varimp_2[0].shape == (3, 4)
assert len(varimp_2[1]) == 4
assert len(varimp_2[2]) == 3
assert isinstance(aml.varimp_heatmap().figure(), matplotlib.pyplot.Figure)
assert isinstance(h2o.varimp_heatmap(aml.leaderboard[aml.leaderboard["model_id"].grep("Stacked", invert=True, output_logical=True), :].head(3), num_of_features=3).figure(), matplotlib.pyplot.Figure)
def test_explanation_list_of_models_binomial_classification():
train = h2o.upload_file(
pyunit_utils.locate("smalldata/logreg/prostate.csv"))
y = "CAPSULE"
train[y] = train[y].asfactor()
# get at most one column from each type
cols_to_test = []
for col, typ in train.types.items():
for ctt in cols_to_test:
if typ == train.types[ctt] or col == y:
break
else:
cols_to_test.append(col)
aml = H2OAutoML(seed=1234, max_models=5)
aml.train(y=y, training_frame=train)
models = [
h2o.get_model(m[0])
for m in aml.leaderboard["model_id"].as_data_frame(use_pandas=False,
header=False)
]
# Test named models as well
gbm = H2OGradientBoostingEstimator(model_id="my_awesome_model")
gbm.train(y=y, training_frame=train)
models += [gbm]
# test variable importance heatmap plot
assert isinstance(
h2o.varimp_heatmap(models).figure(), matplotlib.pyplot.Figure)
matplotlib.pyplot.close()
# test model correlation heatmap plot
assert isinstance(
h2o.model_correlation_heatmap(models, train).figure(),
matplotlib.pyplot.Figure)
matplotlib.pyplot.close()
# test partial dependences
for col in cols_to_test:
assert isinstance(
h2o.pd_multi_plot(models, train, col).figure(),
matplotlib.pyplot.Figure)
matplotlib.pyplot.close()
# test learning curve
for model in models:
assert isinstance(model.learning_curve_plot().figure(),
matplotlib.pyplot.Figure)
matplotlib.pyplot.close("all")
# test explain
assert isinstance(h2o.explain(models, train, render=False), H2OExplanation)
# test explain row
assert isinstance(h2o.explain_row(models, train, 1, render=False),
H2OExplanation)
def test_varimp_heatmap_model_correlation_heatmap():
train = h2o.upload_file(
pyunit_utils.locate("smalldata/logreg/prostate.csv"))
y = "CAPSULE"
train[y] = train[y].asfactor()
# get at most one column from each type
cols_to_test = []
for col, typ in train.types.items():
for ctt in cols_to_test:
if typ == train.types[ctt] or col == y:
break
else:
cols_to_test.append(col)
aml = H2OAutoML(seed=1234, max_models=5)
aml.train(y=y, training_frame=train)
models = [
h2o.get_model(m[0])
for m in aml.leaderboard["model_id"].as_data_frame(use_pandas=False,
header=False)
]
# Test named models as well
gbm = H2OGradientBoostingEstimator(model_id="my_awesome_model")
gbm.train(y=y, training_frame=train)
models += [gbm]
with TemporaryDirectory() as tmpdir:
path1 = "{}/plot1.png".format(tmpdir)
path2 = "{}/plot2.png".format(tmpdir)
test_plot_result_saving(
h2o.varimp_heatmap(models), path2,
h2o.varimp_heatmap(models, save_plot_path=path1), path1)
test_plot_result_saving(
h2o.model_correlation_heatmap(models, train), path2,
h2o.model_correlation_heatmap(models, train, save_plot_path=path1),
path1)
h2o.varimp_heatmap(models)
def test_explanation_list_of_models_multinomial_classification():
train = h2o.upload_file(pyunit_utils.locate("smalldata/iris/iris2.csv"))
y = "response"
train[y] = train[y].asfactor()
# get at most one column from each type
cols_to_test = []
for col, typ in train.types.items():
for ctt in cols_to_test:
if typ == train.types[ctt] or col == y:
break
else:
cols_to_test.append(col)
aml = H2OAutoML(seed=1234, max_models=5)
aml.train(y=y, training_frame=train)
models = [h2o.get_model(m[0]) for m in
aml.leaderboard["model_id"].as_data_frame(use_pandas=False, header=False)]
# test variable importance heatmap plot
assert isinstance(h2o.varimp_heatmap(models), matplotlib.pyplot.Figure)
matplotlib.pyplot.close()
# test model correlation heatmap plot
assert isinstance(h2o.model_correlation_heatmap(models, train), matplotlib.pyplot.Figure)
matplotlib.pyplot.close()
# test partial dependences
for col in cols_to_test:
assert isinstance(h2o.pd_multi_plot(models, train, col, target="setosa"), matplotlib.pyplot.Figure)
matplotlib.pyplot.close("all")
# test explain
assert isinstance(h2o.explain(models, train, render=False), H2OExplanation)
# test explain row
assert isinstance(h2o.explain_row(models, train, 1, render=False), H2OExplanation)