def stackedensemble_metalearner_seed_test():
# Import training set
train = h2o.import_file(path=pyunit_utils.locate("smalldata/testng/higgs_train_5k.csv"),
destination_frame="higgs_train_5k")
test = h2o.import_file(path=pyunit_utils.locate("smalldata/testng/higgs_test_5k.csv"),
destination_frame="higgs_test_5k")
# Identify predictors and response
x = train.columns
y = "response"
x.remove(y)
# Convert response to a factor
train[y] = train[y].asfactor()
test[y] = test[y].asfactor()
# Set number of folds for base learners
nfolds = 3
#Metalearner params for gbm, drf, glm, and deep deeplearning
gbm_params = {"sample_rate" : 0.3, "col_sample_rate" : 0.3}
# Train and cross-validate a GBM
my_gbm = H2OGradientBoostingEstimator(distribution="bernoulli",
ntrees=10,
nfolds=nfolds,
keep_cross_validation_predictions=True,
seed=1)
my_gbm.train(x=x, y=y, training_frame=train)
# Train and cross-validate a RF
my_rf = H2ORandomForestEstimator(ntrees=10,
nfolds=nfolds,
keep_cross_validation_predictions=True,
seed=1)
my_rf.train(x=x, y=y, training_frame=train)
#Train two SE models with same metalearner seeds
stack_gbm1 = H2OStackedEnsembleEstimator(base_models=[my_gbm, my_rf], metalearner_algorithm="gbm",
metalearner_params = gbm_params, seed = 55555)
stack_gbm2 = H2OStackedEnsembleEstimator(base_models=[my_gbm, my_rf], metalearner_algorithm="gbm",
metalearner_params = gbm_params, seed = 55555)
stack_gbm1.train(x=x, y=y, training_frame=train)
stack_gbm2.train(x=x, y=y, training_frame=train)
meta_gbm1 = h2o.get_model(stack_gbm1.metalearner()['name'])
meta_gbm2 = h2o.get_model(stack_gbm2.metalearner()['name'])
assert meta_gbm1.rmse(train=True) == meta_gbm2.rmse(train=True), "RMSE should match if same seed"
#Train two SE models with diff metalearner seeds
stack_gbm3 = H2OStackedEnsembleEstimator(base_models=[my_gbm, my_rf], metalearner_algorithm="gbm",
metalearner_params = gbm_params, seed = 55555)
stack_gbm4 = H2OStackedEnsembleEstimator(base_models=[my_gbm, my_rf], metalearner_algorithm="gbm",
metalearner_params = gbm_params, seed = 98765)
stack_gbm3.train(x=x, y=y, training_frame=train)
stack_gbm4.train(x=x, y=y, training_frame=train)
meta_gbm3 = h2o.get_model(stack_gbm3.metalearner()['name'])
meta_gbm4 = h2o.get_model(stack_gbm4.metalearner()['name'])
assert meta_gbm3.rmse(train=True) != meta_gbm4.rmse(train=True), "RMSE should NOT match if diff seed"
def covtype_get_model():
covtype = h2o.import_file(path=pyunit_utils.locate("smalldata/covtype/covtype.20k.data"))
Y = 54
X = range(0,20) + range(29,54)
# Set response to be indicator of a particular class
res_class = random.randint(1,4)
# Log.info(paste("Setting response column", myY, "to be indicator of class", res_class, "\n"))
covtype[54] = (covtype[54] == res_class)
# L2: alpha = 0, lambda = 0
covtype_mod1 = H2OGeneralizedLinearEstimator(family="binomial", alpha=0, Lambda=0)
covtype_mod1.train(x=X,y=Y, training_frame=covtype)
covtype_mod1.show()
covtype_mod1 = h2o.get_model(covtype_mod1.model_id)
covtype_mod1.show()
# Elastic: alpha = 0.5, lambda = 1e-4
covtype_mod2 = H2OGeneralizedLinearEstimator(family="binomial", alpha=0.5, Lambda=1e-4)
covtype_mod2.train(x=X, y=Y, training_frame=covtype)
covtype_mod2.show()
covtype_mod2 = h2o.get_model(covtype_mod2.model_id)
covtype_mod2.show()
# L1: alpha = 1, lambda = 1e-4
covtype_mod3 = H2OGeneralizedLinearEstimator(family="binomial", alpha=1, Lambda=1e-4)
covtype_mod3.train(x=X,y=Y, training_frame=covtype)
covtype_mod3.show()
covtype_mod3 = h2o.get_model(covtype_mod3.model_id)
covtype_mod3.show()
def covtype_get_model(ip,port):
#Log.info("Importing covtype.20k.data...\n")
covtype = h2o.import_file(path=h2o.locate("smalldata/covtype/covtype.20k.data"))
Y = 54
X = range(0,20) + range(29,54)
# Set response to be indicator of a particular class
res_class = random.randint(1,4)
# Log.info(paste("Setting response column", myY, "to be indicator of class", res_class, "\n"))
covtype[54] = (covtype[54] == res_class)
#covtype_data.summary()
# L2: alpha = 0, lambda = 0
covtype_mod1 = h2o.glm(y=covtype[Y], x=covtype[X], family="binomial", alpha=[0], Lambda=[0])
covtype_mod1.show()
covtype_mod1 = h2o.get_model(covtype_mod1._id)
covtype_mod1.show()
# Elastic: alpha = 0.5, lambda = 1e-4
covtype_mod2 = h2o.glm(y=covtype[Y], x=covtype[X], family="binomial", alpha=[0.5], Lambda=[1e-4])
covtype_mod2.show()
covtype_mod2 = h2o.get_model(covtype_mod2._id)
covtype_mod2.show()
# L1: alpha = 1, lambda = 1e-4
covtype_mod3 = h2o.glm(y=covtype[Y], x=covtype[X], family="binomial", alpha=[1], Lambda=[1e-4])
covtype_mod3.show()
covtype_mod3 = h2o.get_model(covtype_mod3._id)
covtype_mod3.show()
def get_model_test():
prostate = h2o.import_file(path=pyunit_utils.locate("smalldata/logreg/prostate.csv"))
r = prostate[0].runif()
train = prostate[r < 0.70]
test = prostate[r >= 0.70]
# Regression
regression_gbm1 = H2OGradientBoostingEstimator(distribution="gaussian")
regression_gbm1.train(x=[2,3,4,5,6,7,8], y=1, training_frame=train)
predictions1 = regression_gbm1.predict(test)
regression_gbm2 = h2o.get_model(regression_gbm1._id)
assert regression_gbm2._model_json['output']['model_category'] == "Regression"
predictions2 = regression_gbm2.predict(test)
for r in range(predictions1.nrow):
p1 = predictions1[r,0]
p2 = predictions2[r,0]
assert p1 == p2, "expected regression predictions to be the same for row {}, but got {} and {}".format(r, p1, p2)
# Binomial
train[1] = train[1].asfactor()
bernoulli_gbm1 = H2OGradientBoostingEstimator(distribution="bernoulli")
bernoulli_gbm1.train(x=[2,3,4,5,6,7,8],y=1,training_frame=train)
predictions1 = bernoulli_gbm1.predict(test)
bernoulli_gbm2 = h2o.get_model(bernoulli_gbm1._id)
assert bernoulli_gbm2._model_json['output']['model_category'] == "Binomial"
predictions2 = bernoulli_gbm2.predict(test)
for r in range(predictions1.nrow):
p1 = predictions1[r,0]
p2 = predictions2[r,0]
assert p1 == p2, "expected binomial predictions to be the same for row {}, but got {} and {}".format(r, p1, p2)
# Clustering
benign_h2o = h2o.import_file(path=pyunit_utils.locate("smalldata/logreg/benign.csv"))
km_h2o = H2OKMeansEstimator(k=3)
km_h2o.train(x=list(range(benign_h2o.ncol)), training_frame=benign_h2o)
benign_km = h2o.get_model(km_h2o._id)
assert benign_km._model_json['output']['model_category'] == "Clustering"
# Multinomial
train[4] = train[4].asfactor()
multinomial_dl1 = H2ODeepLearningEstimator(loss="CrossEntropy")
multinomial_dl1.train(x=[0,1], y=4, training_frame=train)
predictions1 = multinomial_dl1.predict(test)
multinomial_dl2 = h2o.get_model(multinomial_dl1._id)
assert multinomial_dl2._model_json['output']['model_category'] == "Multinomial"
predictions2 = multinomial_dl2.predict(test)
for r in range(predictions1.nrow):
p1 = predictions1[r,0]
p2 = predictions2[r,0]
assert p1 == p2, "expected multinomial predictions to be the same for row {0}, but got {1} and {2}" \
"".format(r, p1, p2)
def get_model_test(ip,port):
# Connect to h2o
h2o.init(ip,port)
prostate = h2o.import_frame(path=h2o.locate("smalldata/logreg/prostate.csv"))
r = prostate[0].runif()
train = prostate[r < 0.70]
test = prostate[r >= 0.30]
# Regression
regression_gbm1 = h2o.gbm(y=train[1], x=train[2:9], distribution="gaussian")
predictions1 = regression_gbm1.predict(test)
regression_gbm2 = h2o.get_model(regression_gbm1._key)
assert regression_gbm2._model_json['output']['model_category'] == "Regression"
predictions2 = regression_gbm2.predict(test)
for r in range(predictions1.nrow()):
p1 = predictions1[r,0]
p2 = predictions2[r,0]
assert p1 == p2, "expected regression predictions to be the same for row {0}, but got {1} and {2}" \
"".format(r, p1, p2)
# Binomial
train[1] = train[1].asfactor()
bernoulli_gbm1 = h2o.gbm(y=train[1], x=train[2:9], distribution="bernoulli")
predictions1 = bernoulli_gbm1.predict(test)
bernoulli_gbm2 = h2o.get_model(bernoulli_gbm1._key)
assert bernoulli_gbm2._model_json['output']['model_category'] == "Binomial"
predictions2 = bernoulli_gbm2.predict(test)
for r in range(predictions1.nrow()):
p1 = predictions1[r,0]
p2 = predictions2[r,0]
assert p1 == p2, "expected binomial predictions to be the same for row {0}, but got {1} and {2}" \
"".format(r, p1, p2)
# Clustering
benign_h2o = h2o.import_frame(path=h2o.locate("smalldata/logreg/benign.csv"))
km_h2o = h2o.kmeans(x=benign_h2o, k=3)
benign_km = h2o.get_model(km_h2o._key)
assert benign_km._model_json['output']['model_category'] == "Clustering"
# Multinomial
train[4] = train[4].asfactor()
multinomial_dl1 = h2o.deeplearning(x=train[0:2], y=train[4], loss='CrossEntropy')
predictions1 = multinomial_dl1.predict(test)
multinomial_dl2 = h2o.get_model(multinomial_dl1._key)
assert multinomial_dl2._model_json['output']['model_category'] == "Multinomial"
predictions2 = multinomial_dl2.predict(test)
for r in range(predictions1.nrow()):
p1 = predictions1[r,0]
p2 = predictions2[r,0]
assert p1 == p2, "expected multinomial predictions to be the same for row {0}, but got {1} and {2}" \
"".format(r, p1, p2)
def get_xval_models(self, key=None):
"""
Return a Model object.
:param key: If None, return all cross-validated models; otherwise return the model that key points to.
:return: A model or list of models.
"""
return h2o.get_model(key) if key is not None else [h2o.get_model(k) for k in self._xval_keys]
def test_param_disabled():
print("\n=== disabling "+kcvp+" ===")
aml = setup_and_train(False)
_, non_se, se = get_partitioned_model_names(aml.leaderboard)
keys = list_keys_in_memory()
preds = len(keys['cv_predictions'])
assert preds == 0, "{preds} CV predictions were not cleaned from memory".format(preds=preds)
for m in non_se:
assert_cv_predictions_on_model(m, False)
for m in se:
assert not h2o.get_model(h2o.get_model(m).metalearner()['name']).cross_validation_predictions()
def get_grid(self, sort_by=None, decreasing=None):
"""
Retrieve an H2OGridSearch instance. Optionally specify a metric by which to sort models and a sort order.
Note that if neither cross-validation nor a validation frame is used in the grid search, then the
training metrics will display in the "get grid" output. If a validation frame is passed to the grid, and
``nfolds = 0``, then the validation metrics will display. However, if ``nfolds`` > 1, then cross-validation
metrics will display even if a validation frame is provided.
Parameters
----------
sort_by : str, optional
A metric by which to sort the models in the grid space. Choices are "logloss", "residual_deviance", "mse",
"auc", "r2", "accuracy", "precision", "recall", "f1", etc.
decreasing : bool, optional
Sort the models in decreasing order of metric if true, otherwise sort in increasing order (default).
Returns
-------
A new H2OGridSearch instance optionally sorted on the specified metric.
"""
if sort_by is None and decreasing is None: return self
grid_json = h2o.api("GET /99/Grids/%s" % self._id, data={"sort_by": sort_by, "decreasing": decreasing})
grid = H2OGridSearch(self.model, self.hyper_params, self._id)
grid.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']] # reordered
first_model_json = h2o.api("GET /99/Models/%s" % grid_json['model_ids'][0]['name'])['models'][0]
model_class = H2OGridSearch._metrics_class(first_model_json)
m = model_class()
m._id = self._id
m._grid_json = grid_json
# m._metrics_class = metrics_class
m._parms = grid._parms
H2OEstimator.mixin(grid, model_class)
grid.__dict__.update(m.__dict__.copy())
return grid
def predict(self, test_data):
"""
Predict on a dataset.
:param H2OFrame test_data: Data on which to make predictions.
:returns: A new H2OFrame of predictions.
:examples:
>>> #Set up an H2OAutoML object
>>> build_control = {
>>> 'stopping_criteria': {
>>> 'stopping_rounds': 3,
>>> 'stopping_tolerance': 0.001
>>> }
>>> }
>>> aml = H2OAutoML(max_runtime_secs=30, build_control=build_control)
>>> # Launch H2OAutoML
>>> aml.train(y=y, training_frame=training_frame)
>>> #Predict with #1 model from H2OAutoML leaderboard
>>> aml.predict(test_data)
"""
if self._fetch():
self._model = h2o.get_model(self._leader_id)
return self._model.predict(test_data)
print("No model built yet...")
def test_workaround_for_distribution():
try:
h2o.rapids("(setproperty \"{}\" \"{}\")".format(
"sys.ai.h2o.automl.algo_parameters.all.enabled", "true"))
ds = import_dataset('regression')
aml = H2OAutoML(project_name="py_test",
algo_parameters=dict(
distribution='poisson',
family='poisson',
),
exclude_algos=['StackedEnsemble'],
max_runtime_secs=60,
seed=1)
aml.train(y=ds.target, training_frame=ds.train)
model_names = [
aml.leaderboard[i, 0] for i in range(0, (aml.leaderboard.nrows))
]
for mn in model_names:
m = h2o.get_model(mn)
dist = m.params[
'distribution'] if 'distribution' in m.params else m.params[
'family'] if 'family' in m.params else None
print("{}: distribution = {}".format(mn, dist))
except:
h2o.rapids("(setproperty \"{}\" \"{}\")".format(
"sys.ai.h2o.automl.algo_parameters.all.enabled", "false"))
def grid_resume():
train = h2o.import_file(path=pyunit_utils.locate("smalldata/iris/iris_wheader.csv"))
# Run GBM Grid Search
ntrees_opts = [1, 3]
learn_rate_opts = [0.1, .05]
hyper_parameters = OrderedDict()
hyper_parameters["learn_rate"] = learn_rate_opts
hyper_parameters["ntrees"] = ntrees_opts
print("GBM grid with the following hyper_parameters:", hyper_parameters)
export_dir = pyunit_utils.locate("results")
gs = H2OGridSearch(H2OGradientBoostingEstimator, hyper_params=hyper_parameters)
gs.train(x=list(range(4)), y=4, training_frame=train)
grid_id = gs.grid_id
old_grid_model_count = len(gs.model_ids)
print("Baseline grid has %d models" % old_grid_model_count)
saved_path = h2o.save_grid(export_dir, grid_id)
h2o.remove_all()
train = h2o.import_file(path=pyunit_utils.locate("smalldata/iris/iris_wheader.csv"))
grid = h2o.load_grid(saved_path)
assert grid is not None
assert len(grid.model_ids) == old_grid_model_count
# Modify the hyperspace - should add new models to the grid
hyper_parameters["ntrees"] = [2, 5]
grid = H2OGridSearch(H2OGradientBoostingEstimator, hyper_params=hyper_parameters, grid_id = grid.grid_id)
grid.train(x=list(range(4)), y=4, training_frame=train)
print("Newly grained grid has %d models" % len(grid.model_ids))
assert len(grid.model_ids) == 2 * old_grid_model_count
for model_id in grid.model_ids:
model = h2o.get_model(model_id)
assert model is not None
def test_algo_parameter_can_be_applied_only_to_a_specific_algo():
ds = import_dataset()
aml = H2OAutoML(
project_name="py_specific_algo_param",
algo_parameters=dict(GBM__monotone_constraints=dict(AGE=1)),
max_models=6,
seed=1)
aml.train(y=ds.target, training_frame=ds.train)
model_names, _, _ = get_partitioned_model_names(aml.leaderboard)
models_supporting_monotone_constraints = [
n for n in model_names if re.match(r"GBM|XGBoost", n)
]
assert next((m for m in models_supporting_monotone_constraints
if m.startswith('GBM')),
None), "There should be at least one GBM model"
for m in models_supporting_monotone_constraints:
model = h2o.get_model(m)
mc_value = next(v['actual'] for n, v in model.params.items()
if n == 'monotone_constraints')
if m.startswith('GBM'):
assert isinstance(mc_value, list)
age = next((v for v in mc_value if v['key'] == 'AGE'), None)
assert age is not None
assert age['value'] == 1.0
else:
assert mc_value is None
def get_modelKmeans(ip, port):
# Connect to a pre-existing cluster
# connect to localhost:54321
#Log.info("Importing benign.csv data...\n")
benign_h2o = h2o.import_file(
path=h2o.locate("smalldata/logreg/benign.csv"))
#benign_h2o.summary()
benign_sci = np.genfromtxt(h2o.locate("smalldata/logreg/benign.csv"),
delimiter=",")
# Impute missing values with column mean
imp = Imputer(missing_values='NaN', strategy='mean', axis=0)
benign_sci = imp.fit_transform(benign_sci)
for i in range(2, 7):
# Log.info("H2O K-Means")
km_h2o = h2o.kmeans(x=benign_h2o, k=i)
km_h2o.show()
model = h2o.get_model(km_h2o._id)
model.show()
km_sci = KMeans(n_clusters=i, init='k-means++', n_init=1)
km_sci.fit(benign_sci)
print "sckit centers"
print km_sci.cluster_centers_
def test_automl_creates_interpretable_SE_with_only_monotonic_models():
ds = import_dataset()
aml_mono = H2OAutoML(
project_name="test_automl_creates_interpretable_se",
max_models=5,
include_algos=["GBM", "GLM", "XGBoost", "StackedEnsemble"],
monotone_constraints=dict(AGE=1,
DPROS=1,
DCAPS=1,
PSA=1,
VOL=1,
GLEASON=1),
seed=1234)
aml_mono.train(y=ds.target, training_frame=ds.train)
leaderboard = (aml_mono.leaderboard.as_data_frame()["model_id"])
assert leaderboard.apply(
lambda model_name: "Monotonic" in model_name).any()
se_name = leaderboard[leaderboard.apply(
lambda model_name: "Monotonic" in model_name)]
se_mono = h2o.get_model(se_name.iloc[0])
assert leaderboard.apply(lambda model_name: 'GLM' in model_name).any()
assert all(['GBM' in bm or 'XGBoost' in bm for bm in se_mono.base_models])
def test_param_disabled():
print("\n=== disabling "+kcvm+" ===")
aml = setup_and_train(False)
models, non_se, se = get_partitioned_model_names(aml.leaderboard)
check_model_property(se, kcvm, False)
check_model_property(non_se, kcvm, True, False, True)
keys = list_keys_in_memory()
tot, cv = len(keys['models']), len(keys['cv_models'])
print("total models in memory = {tot}, among which {cv} CV models".format(tot=tot, cv=cv))
assert tot > 0, "no models left in memory"
assert cv == 0, "{cv} CV models were not cleaned from memory".format(cv=cv)
for m in non_se:
assert not h2o.get_model(m).cross_validation_models(), "unexpected cv models for model "+m
for m in se:
metal = h2o.get_model(h2o.get_model(m).metalearner()['name'])
assert not metal.cross_validation_models(), "unexpected cv models for metalearner of model "+m
def test_monotone_constraints_can_be_passed_as_algo_parameter():
ds = import_dataset()
aml = H2OAutoML(
project_name="py_monotone_constraints",
algo_parameters=dict(
monotone_constraints=dict(
AGE=1, VOL=-1), # constraints just for the sake of testing
# ntrees=10,
),
max_models=6,
seed=1)
aml.train(y=ds.target, training_frame=ds.train)
model_names, _, _ = get_partitioned_model_names(aml.leaderboard)
models_supporting_monotone_constraints = [
n for n in model_names if re.match(r"GBM|XGBoost", n)
]
assert len(models_supporting_monotone_constraints) < len(model_names), \
"models not supporting the constraint should not have been skipped"
for m in models_supporting_monotone_constraints:
model = h2o.get_model(m)
value = next(v['actual'] for n, v in model.params.items()
if n == 'monotone_constraints')
# print(param)
assert isinstance(value, list)
assert len(value) == 2
age = next((v for v in value if v['key'] == 'AGE'), None)
assert age is not None
assert age['value'] == 1.0
vol = next((v for v in value if v['key'] == 'VOL'), None)
assert vol is not None
assert vol['value'] == -1.0
def _fetch_state(aml_id, properties=None):
state_json = h2o.api("GET /99/AutoML/%s" % aml_id)
project_name = state_json["project_name"]
if project_name is None:
raise H2OValueError("No AutoML instance with id {}.".format(aml_id))
leaderboard_list = [key["name"] for key in state_json['leaderboard']['models']]
leader_id = leaderboard_list[0] if (leaderboard_list is not None and len(leaderboard_list) > 0) else None
should_fetch = lambda prop: properties is None or prop in properties
leader = None
if should_fetch('leader'):
leader = h2o.get_model(leader_id) if leader_id is not None else None
leaderboard = None
if should_fetch('leaderboard'):
leaderboard = H2OAutoML._fetch_table(state_json['leaderboard_table'], key=project_name+"_leaderboard", progress_bar=False)
leaderboard = h2o.assign(leaderboard[1:], project_name+"_leaderboard") # removing index and reassign id to ensure persistence on backend
event_log = None
if should_fetch('event_log'):
event_log = H2OAutoML._fetch_table(state_json['event_log_table'], key=project_name+"_eventlog", progress_bar=False)
event_log = h2o.assign(event_log[1:], project_name+"_eventlog") # removing index and reassign id to ensure persistence on backend
return dict(
project_name=project_name,
json=state_json,
leader_id=leader_id,
leader=leader,
leaderboard=leaderboard,
event_log=event_log,
)
def test_param_disabled():
print("\n=== disabling "+kcvm+" ===")
aml = setup_and_train(False)
models, non_se, se = get_partitioned_model_names(aml.leaderboard)
check_model_property(se, kcvm, False)
check_model_property(non_se, kcvm, True, False, True)
keys = list_keys_in_memory()
tot, cv = len(keys['models_all']), len(keys['cv_models'])
print("total models in memory = {tot}, among which {cv} CV models".format(tot=tot, cv=cv))
assert tot > 0, "no models left in memory"
assert cv == 0, "{cv} CV models were not cleaned from memory".format(cv=cv)
for m in non_se:
assert not h2o.get_model(m).cross_validation_models(), "unexpected cv models for model "+m
for m in se:
metal = h2o.get_model(h2o.get_model(m).metalearner()['name'])
assert not metal.cross_validation_models(), "unexpected cv models for metalearner of model "+m
def get_modelKmeans():
# Connect to a pre-existing cluster
# connect to localhost:54321
# Log.info("Importing benign.csv data...\n")
benign_h2o = h2o.import_file(path=pyunit_utils.locate("smalldata/logreg/benign.csv"))
# benign_h2o.summary()
benign_sci = np.genfromtxt(pyunit_utils.locate("smalldata/logreg/benign.csv"), delimiter=",")
# Impute missing values with column mean
imp = Imputer(missing_values="NaN", strategy="mean", axis=0)
benign_sci = imp.fit_transform(benign_sci)
for i in range(2, 7):
# Log.info("H2O K-Means")
km_h2o = H2OKMeansEstimator(k=i)
km_h2o.train(x=range(benign_h2o.ncol), training_frame=benign_h2o)
km_h2o.show()
model = h2o.get_model(km_h2o._id)
model.show()
km_sci = KMeans(n_clusters=i, init="k-means++", n_init=1)
km_sci.fit(benign_sci)
print "sckit centers"
print km_sci.cluster_centers_
def get_grid(self, sort_by=None, decreasing=None):
"""
Retrieve an H2OGridSearch instance. Optionally specify a metric by which to sort models and a sort order.
Parameters
----------
sort_by : str, optional
A metric by which to sort the models in the grid space. Choices are "logloss", "residual_deviance", "mse", "auc", "r2", "accuracy", "precision", "recall", "f1", etc.
decreasing : bool, optional
Sort the models in decreasing order of metric if true, otherwise sort in increasing order (default).
Returns
-------
A new H2OGridSearch instance optionally sorted on the specified metric.
"""
if sort_by is None and decreasing is None: return self
grid_json = H2OConnection.get_json("Grids/"+self._id, sort_by=sort_by, decreasing=decreasing, _rest_version=99)
grid = H2OGridSearch(self.model, self.hyper_params, self._id)
grid.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']] #reordered
first_model_json = H2OConnection.get_json("Models/"+grid_json['model_ids'][0]['name'], _rest_version=99)['models'][0]
model_class = H2OGridSearch._metrics_class(first_model_json)
m = model_class()
m._id = self._id
m._grid_json = grid_json
# m._metrics_class = metrics_class
m._parms = grid._parms
H2OEstimator.mixin(grid,model_class)
grid.__dict__.update(m.__dict__.copy())
return grid
def test_base_models_are_populated():
train = h2o.import_file(pu.locate("smalldata/iris/iris_train.csv"))
test = h2o.import_file(pu.locate("smalldata/iris/iris_test.csv"))
x = train.columns
y = "species"
x.remove(y)
nfolds = 2
gbm = H2OGradientBoostingEstimator(nfolds=nfolds,
fold_assignment="Modulo",
keep_cross_validation_predictions=True)
gbm.train(x=x, y=y, training_frame=train)
rf = H2ORandomForestEstimator(nfolds=nfolds,
fold_assignment="Modulo",
keep_cross_validation_predictions=True)
rf.train(x=x, y=y, training_frame=train)
se = H2OStackedEnsembleEstimator(training_frame=train,
validation_frame=test,
base_models=[gbm.model_id, rf.model_id])
se.train(x=x, y=y, training_frame=train)
retrieved_se = get_model(se.model_id)
assert len(se.base_models) == 2
assert len(retrieved_se.base_models) == 2
assert se.base_models == retrieved_se.base_models
# ensure that we are getting the model_ids
assert pu.is_type(se.base_models, [str])
assert pu.is_type(retrieved_se.base_models, [str])
def save_model(model_id, dest_dir='.', mformat='mojo'):
model = h2o.get_model(model_id)
if mformat == 'mojo':
model.save_mojo(path=dest_dir)
# model.download_mojo(path=dest_dir, get_genmodel_jar=True)
else:
model.save_model_details(path=dest_dir)
def get_hyperparams_dict(self, id, display=True):
"""
Derived and returned the model parameters used to train the particular grid search model.
Parameters
----------
id: str
The model id of the model with hyperparameters of interest.
display: boolean
Flag to indicate whether to display the hyperparameter names.
Returns
-------
A dict of model pararmeters derived from the hyper-parameters used to train this particular model.
"""
idx = id if is_int(id) else self.model_ids.index(id)
model = self[idx]
model_params = dict()
# if cross-validation is turned on, parameters in one of the fold model actual contains the max_runtime_secs
# parameter and not the main model that is returned.
if model._is_xvalidated:
model = h2o.get_model(model._xval_keys[0])
for param_name in self.hyper_names:
model_params[param_name] = model.params[param_name]['actual'][0] if \
isinstance(model.params[param_name]['actual'], list) else model.params[param_name]['actual']
if display: print('Hyperparameters: [' + ', '.join(list(self.hyper_params.keys())) + ']')
return model_params
def _model_build(self, x, y, tframe, vframe, kwargs):
kwargs['training_frame'] = tframe
if vframe is not None: kwargs["validation_frame"] = vframe
if isinstance(y, int): y = tframe.names[y]
if y is not None: kwargs['response_column'] = y
if not isinstance(x, (list,tuple)): x=[x]
if isinstance(x[0], int):
x = [tframe.names[i] for i in x]
offset = kwargs["offset_column"]
folds = kwargs["fold_column"]
weights= kwargs["weights_column"]
ignored_columns = list(set(tframe.names) - set(x + [y,offset,folds,weights]))
kwargs["ignored_columns"] = None if ignored_columns==[] else [h2o.h2o._quoted(col) for col in ignored_columns]
kwargs = dict([(k, kwargs[k].frame_id if isinstance(kwargs[k], H2OFrame) else kwargs[k]) for k in kwargs if kwargs[k] is not None]) # gruesome one-liner
algo = self.model._compute_algo() #unique to grid search
kwargs["_rest_version"] = 99 #unique to grid search
grid = H2OJob(H2OConnection.post_json("Grid/"+algo, **kwargs), job_type=(algo+" Grid Build"))
if self._future:
self._job = grid
return
grid.poll()
if '_rest_version' in kwargs.keys(): grid_json = H2OConnection.get_json("Grids/"+grid.dest_key, _rest_version=kwargs['_rest_version'])
else: grid_json = H2OConnection.get_json("Grids/"+grid.dest_key)
self.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]
#get first model returned in list of models from grid search to get model class (binomial, multinomial, etc)
first_model_json = H2OConnection.get_json("Models/"+grid_json['model_ids'][0]['name'], _rest_version=kwargs['_rest_version'])['models'][0]
self._resolve_grid(grid.dest_key, grid_json, first_model_json)
def get_grid(self, sort_by=None, decreasing=None):
"""
Retrieve an H2OGridSearch instance. Optionally specify a metric by which to sort models and a sort order.
Parameters
----------
sort_by : str, optional
A metric by which to sort the models in the grid space. Choices are "logloss", "residual_deviance", "mse",
"auc", "r2", "accuracy", "precision", "recall", "f1", etc.
decreasing : bool, optional
Sort the models in decreasing order of metric if true, otherwise sort in increasing order (default).
Returns
-------
A new H2OGridSearch instance optionally sorted on the specified metric.
"""
if sort_by is None and decreasing is None: return self
grid_json = h2o.api("GET /99/Grids/%s" % self._id, data={"sort_by": sort_by, "decreasing": decreasing})
grid = H2OGridSearch(self.model, self.hyper_params, self._id)
grid.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']] # reordered
first_model_json = h2o.api("GET /99/Models/%s" % grid_json['model_ids'][0]['name'])['models'][0]
model_class = H2OGridSearch._metrics_class(first_model_json)
m = model_class()
m._id = self._id
m._grid_json = grid_json
# m._metrics_class = metrics_class
m._parms = grid._parms
H2OEstimator.mixin(grid, model_class)
grid.__dict__.update(m.__dict__.copy())
return grid
def get_hyperparams_dict(self, id, display=True):
"""
Derived and returned the model parameters used to train the particular grid search model.
Parameters
----------
id: str
The model id of the model with hyperparameters of interest.
display: boolean
Flag to indicate whether to display the hyperparameter names.
Returns
-------
A dict of model pararmeters derived from the hyper-parameters used to train this particular model.
"""
idx = id if isinstance(id, int) else self.model_ids.index(id)
model = self[idx]
model_params = dict()
# if cross-validation is turned on, parameters in one of the fold model actual contains the max_runtime_secs
# parameter and not the main model that is returned.
if model._is_xvalidated:
model = h2o.get_model(model._xval_keys[0])
for param_name in self.hyper_names:
model_params[param_name] = model.params[param_name]['actual'][0] if \
isinstance(model.params[param_name]['actual'], list) else model.params[param_name]['actual']
if display:
print('Hyperparameters: [' +
', '.join(list(self.hyper_params.keys())) + ']')
return model_params
def get_hyperparams(self, id, display=True):
"""
Get the hyperparameters of a model explored by grid search.
Parameters
----------
id: str
The model id of the model with hyperparameters of interest.
display: boolean
Flag to indicate whether to display the hyperparameter names.
Returns
-------
A list of the hyperparameters for the specified model.
"""
idx = id if isinstance(id, int) else self.model_ids.index(id)
model = self[idx]
# if cross-validation is turned on, parameters in one of the fold model actuall contains the max_runtime_secs
# parameter and not the main model that is returned.
if model._is_xvalidated:
model = h2o.get_model(model._xval_keys[0])
res = [
model.params[h]['actual'][0] if isinstance(
model.params[h]['actual'], list) else model.params[h]['actual']
for h in self.hyper_params
]
if display:
print('Hyperparameters: [' +
', '.join(list(self.hyper_params.keys())) + ']')
return res
def test_stacked_ensembles_are_trained_with_blending_frame_even_if_nfolds_eq_0(
):
print(
"Check that we can disable cross-validation when passing a blending frame and that Stacked Ensembles are trained using this frame."
)
max_models = 5
ds = import_dataset()
aml = H2OAutoML(project_name="py_aml_blending_frame",
seed=1,
max_models=max_models,
nfolds=0)
aml.train(y=ds.target,
training_frame=ds.train,
blending_frame=ds.valid,
leaderboard_frame=ds.test)
se = get_partitioned_model_names(aml.leaderboard).se
assert len(
se
) > 3, "In blending mode, StackedEnsemble should still be trained in spite of nfolds=0."
for m in se:
model = h2o.get_model(m)
assert model.params['blending_frame']['actual'][
'name'] == ds.valid.frame_id
assert model._model_json['output']['stacking_strategy'] == 'blending'
def test_stackedensemble_propagates_the_max_runtime_secs():
max_runtime_secs = 5
hyper_parameters = dict()
hyper_parameters["ntrees"] = [1, 3, 5]
params = dict(
fold_assignment="modulo",
nfolds=3,
keep_cross_validation_predictions=True
)
data = prepare_data()
gs1 = H2OGridSearch(H2OGradientBoostingEstimator(**params), hyper_params=hyper_parameters)
gs1.train(data.x, data.y, data.train, validation_frame=data.train)
se = H2OStackedEnsembleEstimator(base_models=[gs1], max_runtime_secs=max_runtime_secs)
se.train(data.x, data.y, data.train)
metalearner = h2o.get_model(se.metalearner()["name"])
# metalearner has the set max_runtine_secs
assert metalearner.actual_params['max_runtime_secs'] <= max_runtime_secs
assert metalearner.actual_params['max_runtime_secs'] > 0
# stack ensemble has the set max_runtime_secs
assert se.max_runtime_secs == max_runtime_secs
def get_automl(project_name):
"""
Retrieve information about an AutoML instance.
:param str project_name: A string indicating the project_name of the automl instance to retrieve.
:returns: A dictionary containing the project_name, leader model, and leaderboard.
"""
automl_json = h2o.api("GET /99/AutoML/%s" % project_name)
project_name = automl_json["project_name"]
leaderboard_list = [key["name"] for key in automl_json['leaderboard']['models']]
if leaderboard_list is not None and len(leaderboard_list) > 0:
leader_id = leaderboard_list[0]
else:
leader_id = None
leader = h2o.get_model(leader_id)
# Intentionally mask the progress bar here since showing multiple progress bars is confusing to users.
# If any failure happens, revert back to user's original setting for progress and display the error message.
is_progress = H2OJob.__PROGRESS_BAR__
h2o.no_progress()
try:
# Parse leaderboard H2OTwoDimTable & return as an H2OFrame
leaderboard = h2o.H2OFrame(
automl_json["leaderboard_table"].cell_values,
column_names=automl_json["leaderboard_table"].col_header)
except Exception as ex:
raise ex
finally:
if is_progress is True:
h2o.show_progress()
leaderboard = leaderboard[1:]
automl_dict = {'project_name': project_name, "leader": leader, "leaderboard": leaderboard}
return automl_dict
def test_api_timestamp():
prostate_train = h2o.import_file(path=pyunit_utils.locate("smalldata/logreg/prostate_train.csv"))
prostate_train["CAPSULE"] = prostate_train["CAPSULE"].asfactor()
ntrees = 1
learning_rate = 0.1
depth = 5
min_rows = 10
# Build H2O GBM classification model:
gbm_h2o = H2OGradientBoostingEstimator(ntrees=ntrees, learn_rate=learning_rate,
max_depth=depth,
min_rows=min_rows,
distribution="bernoulli",
model_id="test_timestamp")
gbm_h2o.train(x=list(range(1,prostate_train.ncol)),y="CAPSULE", training_frame=prostate_train)
model = h2o.get_model(model_id="test_timestamp")
models = h2o.api("GET /3/Models")
assert model._model_json['timestamp'] == models["models"][0]["timestamp"], "Timestamp should be the same."
assert gbm_h2o.start_time is not None and gbm_h2o.start_time > 0
assert gbm_h2o.end_time is not None and gbm_h2o.end_time > 0
assert gbm_h2o.run_time is not None and gbm_h2o.run_time > 0
assert gbm_h2o.end_time - gbm_h2o.start_time == gbm_h2o.run_time
def test_maxrglm_gaussian_coefs():
d = h2o.import_file(
path=pyunit_utils.locate("smalldata/logreg/prostate.csv"))
my_y = "GLEASON"
my_x = ["AGE", "RACE", "CAPSULE", "DCAPS", "PSA", "VOL", "DPROS"]
maxrglm_model = maxrglm(seed=12345, max_predictor_number=7)
maxrglm_model.train(training_frame=d, x=my_x, y=my_y)
coefs = maxrglm_model.coef()
coefs_norm = maxrglm_model.coef_norm()
for ind in list(range(len(coefs))):
one_coef = coefs[ind]
one_coef_norm = coefs_norm[ind]
# coefficients obtained from accessing model_id, generate model and access the model coeffs
one_model = h2o.get_model(
maxrglm_model._model_json["output"]["best_model_ids"][ind]['name'])
model_coef = one_model.coef()
model_coef_norm = one_model.coef_norm()
# get coefficients of individual predictor subset size
subset_size = ind + 1
one_model_coef = maxrglm_model.coef(subset_size)
one_model_coef_norm = maxrglm_model.coef_norm(subset_size)
# check coefficient dicts are equal
pyunit_utils.assertCoefDictEqual(one_coef, model_coef, 1e-6)
pyunit_utils.assertCoefDictEqual(one_coef_norm, model_coef_norm, 1e-6)
pyunit_utils.assertCoefDictEqual(one_model_coef, model_coef, 1e-6)
pyunit_utils.assertCoefDictEqual(one_model_coef_norm, model_coef_norm,
1e-6)
def get_grid(model, hyper_params, grid_id):
"""
Retrieve an H2OGridSearch instance already trained given its original model, hyper_params, and grid_id.
Parameters
----------
model : H2O Estimator model
The type of model explored that is initalized with optional parameters which are unchanged across explored models.
hyper_params: dict
A dictionary of string parameters (keys) and a list of values explored by grid search (values).
grid_id : str, optional
The unique id assigned to the grid object.
Returns
-------
A new H2OGridSearch instance that is a replica of the H2OGridSearch instance with the specified grid_id.
"""
kwargs = {'_rest_version':99}
grid_json = H2OConnection.get_json("Grids/"+grid_id, **kwargs)
grid = H2OGridSearch(model, hyper_params, grid_id)
grid.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]
first_model_json = H2OConnection.get_json("Models/"+grid_json['model_ids'][0]['name'], _rest_version=kwargs['_rest_version'])['models'][0]
model_class = H2OGridSearch._metrics_class(first_model_json)
m = model_class()
m._id = grid_id
m._grid_json = grid_json
# m._metrics_class = metrics_class
m._parms = grid._parms
H2OEstimator.mixin(grid,model_class)
grid.__dict__.update(m.__dict__.copy())
return grid
def get_grid(self, sort_by=None, decreasing=None):
"""
Retrieve an H2OGridSearch instance.
Optionally specify a metric by which to sort models and a sort order.
Note that if neither cross-validation nor a validation frame is used in the grid search, then the
training metrics will display in the "get grid" output. If a validation frame is passed to the grid, and
``nfolds = 0``, then the validation metrics will display. However, if ``nfolds`` > 1, then cross-validation
metrics will display even if a validation frame is provided.
:param str sort_by: A metric by which to sort the models in the grid space. Choices are: ``"logloss"``,
``"residual_deviance"``, ``"mse"``, ``"auc"``, ``"r2"``, ``"accuracy"``, ``"precision"``, ``"recall"``,
``"f1"``, etc.
:param bool decreasing: Sort the models in decreasing order of metric if true, otherwise sort in increasing
order (default).
:returns: A new H2OGridSearch instance optionally sorted on the specified metric.
"""
if sort_by is None and decreasing is None: return self
grid_json = h2o.api("GET /99/Grids/%s" % self._id, data={"sort_by": sort_by, "decreasing": decreasing})
grid = H2OGridSearch(self.model, self.hyper_params, self._id)
grid.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']] # reordered
first_model_json = h2o.api("GET /99/Models/%s" % grid_json['model_ids'][0]['name'])['models'][0]
model_class = H2OGridSearch._metrics_class(first_model_json)
m = model_class()
m._id = self._id
m._grid_json = grid_json
# m._metrics_class = metrics_class
m._parms = grid._parms
H2OEstimator.mixin(grid, model_class)
grid.__dict__.update(m.__dict__.copy())
return grid
def get_hyperparams(self, id, display=True):
"""
Get the hyperparameters of a model explored by grid search.
Parameters
----------
id: str
The model id of the model with hyperparameters of interest.
display: boolean
Flag to indicate whether to display the hyperparameter names.
Returns
-------
A list of the hyperparameters for the specified model.
"""
idx = id if is_int(id) else self.model_ids.index(id)
model = self[idx]
# if cross-validation is turned on, parameters in one of the fold model actuall contains the max_runtime_secs
# parameter and not the main model that is returned.
if model._is_xvalidated:
model = h2o.get_model(model._xval_keys[0])
res = [model.params[h]['actual'][0] if isinstance(model.params[h]['actual'], list)
else model.params[h]['actual']
for h in self.hyper_params]
if display: print('Hyperparameters: [' + ', '.join(list(self.hyper_params.keys())) + ']')
return res
def test_maxrglm_gaussian():
d = h2o.import_file(
path=pyunit_utils.locate("smalldata/logreg/prostate.csv"))
my_y = "GLEASON"
my_x = ["AGE", "RACE", "CAPSULE", "DCAPS", "PSA", "VOL", "DPROS"]
maxrglm_model = maxrglm(seed=12345, max_predictor_number=7)
maxrglm_model.train(training_frame=d, x=my_x, y=my_y)
resultFrame = maxrglm_model.result()
numRows = resultFrame.nrows
best_r2_value = maxrglm_model.get_best_R2_values()
for ind in list(range(numRows)):
# r2 from attributes
best_r2 = best_r2_value[ind]
one_model = h2o.get_model(resultFrame["model_id"][ind, 0])
pred = one_model.predict(d)
print("last element of predictor frame: {0}".format(
pred[pred.nrows - 1, pred.ncols - 1]))
assert pred.nrows == d.nrows, "expected dataset row: {0}, actual dataset row: {1}".format(
pred.nrows, d.nrows)
# r2 from result frame
frame_r2 = resultFrame["best_r2_value"][ind, 0]
# r2 from model
model_r2 = one_model.r2()
# make sure all r2 are equal
assert abs(
best_r2 - frame_r2
) < 1e-6, "expected best r2: {0}, actual best r2: {1}".format(
best_r2, frame_r2)
assert abs(
frame_r2 - model_r2
) < 1e-6, "expected best r2: {0}, actual best r2: {1}".format(
model_r2, frame_r2)
def benign_grid():
training_data = h2o.import_file(pyunit_utils.locate("smalldata/logreg/benign.csv"))
Y = 3
X = list(range(3)) + list(range(4,11))
# NOTE: this tests bad parameter value handling; 'a' is not a float:
hyper_parameters = {'alpha': [0.01,0.5,'a'], 'lambda': [1e-5,1e-6]}
gs = H2OGridSearch(H2OGeneralizedLinearEstimator(family='binomial'), hyper_parameters)
gs.train(x=X,y=Y, training_frame=training_data)
for model in gs:
assert isinstance(model, H2OGeneralizedLinearEstimator)
gs.show()
print(gs.sort_by('F1', False))
best_model_id = gs.sort_by('F1', False)['Model Id'][0]
best_model = h2o.get_model(best_model_id)
best_model.predict(training_data)
gs.predict(training_data)
print(gs.get_hyperparams(best_model_id))
print(gs.grid_id)
assert best_model.params['family']['actual'] == 'binomial'
# test search_criteria plumbing
search_criteria = { 'strategy': "RandomDiscrete", 'max_models': 3 }
max_models_g = H2OGridSearch(H2OGeneralizedLinearEstimator(family='binomial'), hyper_parameters, search_criteria=search_criteria)
max_models_g.train(x=X,y=Y, training_frame=training_data)
max_models_g.show()
print(max_models_g.grid_id)
print(max_models_g.sort_by('F1', False))
assert len(max_models_g.models) == 3, "expected 3 models, got: {}".format(len(max_models_g.models))
print(max_models_g.sorted_metric_table())
print(max_models_g.get_grid("r2"))
def benign_grid():
training_data = h2o.import_file(pyunit_utils.locate("smalldata/logreg/benign.csv"))
Y = 3
X = range(3) + range(4,11)
hyper_parameters = {'alpha': [0.01,0.5,'a'], 'lambda': [1e-5,1e-6]}
gs = H2OGridSearch(H2OGeneralizedLinearEstimator(family='binomial'), hyper_parameters)
gs.train(x=X,y=Y, training_frame=training_data)
gs.show()
print gs.sort_by('F1', False)
best_model_id = gs.sort_by('F1', False)['Model Id'][0]
best_model = h2o.get_model(best_model_id)
best_model.predict(training_data)
gs.predict(training_data)
print gs.get_hyperparams(best_model_id)
print gs.grid_id
new_g = H2OGridSearch.get_grid(H2OGeneralizedLinearEstimator(family='binomial'), hyper_parameters, gs.grid_id)
new_g.show()
print new_g.grid_id
print new_g.sort_by('F1', False)
assert best_model.params['family']['actual'] == 'binomial'
def covtype_get_model(ip, port):
# Connect to h2o
h2o.init(ip, port)
#Log.info("Importing covtype.20k.data...\n")
covtype = h2o.import_frame(
path=h2o.locate("smalldata/covtype/covtype.20k.data"))
Y = 54
X = range(0, 20) + range(29, 54)
# Set response to be indicator of a particular class
res_class = random.randint(1, 4)
# Log.info(paste("Setting response column", myY, "to be indicator of class", res_class, "\n"))
covtype[54] = (covtype[54] == res_class)
#covtype_data.summary()
# L2: alpha = 0, lambda = 0
covtype_mod1 = h2o.glm(y=covtype[Y],
x=covtype[X],
family="binomial",
alpha=[0],
Lambda=[0])
covtype_mod1.show()
covtype_mod1 = h2o.get_model(covtype_mod1._id)
covtype_mod1.show()
# Elastic: alpha = 0.5, lambda = 1e-4
covtype_mod2 = h2o.glm(y=covtype[Y],
x=covtype[X],
family="binomial",
alpha=[0.5],
Lambda=[1e-4])
covtype_mod2.show()
covtype_mod2 = h2o.get_model(covtype_mod2._id)
covtype_mod2.show()
# L1: alpha = 1, lambda = 1e-4
covtype_mod3 = h2o.glm(y=covtype[Y],
x=covtype[X],
family="binomial",
alpha=[1],
Lambda=[1e-4])
covtype_mod3.show()
covtype_mod3 = h2o.get_model(covtype_mod3._id)
covtype_mod3.show()
def test_gaussian_result_frame_model_id():
d = h2o.import_file(
path=pyunit_utils.locate("smalldata/logreg/prostate.csv"))
my_y = "GLEASON"
my_x = ["AGE", "RACE", "CAPSULE", "DCAPS", "PSA", "VOL", "DPROS"]
maxr_model = modelSelection(seed=12345,
max_predictor_number=7,
mode="maxr")
maxr_model.train(training_frame=d, x=my_x, y=my_y)
allsubsets_model = modelSelection(seed=12345,
max_predictor_number=7,
mode="allsubsets")
allsubsets_model.train(training_frame=d, x=my_x, y=my_y)
result_frame_allsubsets = allsubsets_model.result()
numRows = result_frame_allsubsets.nrows
best_r2_allsubsets = allsubsets_model.get_best_R2_values()
result_frame_maxr = maxr_model.result()
best_r2_maxr = maxr_model.get_best_R2_values()
for ind in list(range(numRows)):
# r2 from attributes
best_r2_value_allsubsets = best_r2_allsubsets[ind]
one_model_allsubsets = h2o.get_model(
result_frame_allsubsets["model_id"][ind, 0])
pred_allsubsets = one_model_allsubsets.predict(d)
print("last element of predictor frame: {0}".format(
pred_allsubsets[pred_allsubsets.nrows - 1,
pred_allsubsets.ncols - 1]))
assert pred_allsubsets.nrows == d.nrows, "expected dataset row: {0}, actual dataset row: " \
"{1}".format(pred_allsubsets.nrows, d.nrows)
best_r2_value_maxr = best_r2_maxr[ind]
one_model_maxr = h2o.get_model(result_frame_maxr["model_id"][ind, 0])
pred_maxr = one_model_maxr.predict(d)
pyunit_utils.compare_frames_local(
pred_maxr, pred_allsubsets, prob=1,
tol=1e-6) # compare allsubsets and maxr results
# r2 from result frame
frame_r2_allsubsets = result_frame_allsubsets["best_r2_value"][ind, 0]
# r2 from model
model_r2_allsubsets = one_model_allsubsets.r2()
# make sure all r2 are equal
assert abs(best_r2_value_allsubsets-frame_r2_allsubsets) < 1e-6, "expected best r2: {0}, actual best r2: " \
"{1}".format(best_r2_value_allsubsets, frame_r2_allsubsets)
assert abs(frame_r2_allsubsets-model_r2_allsubsets) < 1e-6, "expected best r2: {0}, actual best r2: " \
"{1}".format(model_r2_allsubsets, frame_r2_allsubsets)
assert abs(best_r2_value_maxr-model_r2_allsubsets) < 1e-6, "expected best r2: {0}, maxr best r2: {1}" \
"".format(best_r2_value_maxr, model_r2_allsubsets)
def _model_build(self, x, y, tframe, vframe, kwargs):
kwargs['training_frame'] = tframe
if vframe is not None: kwargs["validation_frame"] = vframe
if is_type(y, int): y = tframe.names[y]
if y is not None: kwargs['response_column'] = y
if not is_type(x, list, tuple): x = [x]
if is_type(x[0], int):
x = [tframe.names[i] for i in x]
offset = kwargs["offset_column"]
folds = kwargs["fold_column"]
weights = kwargs["weights_column"]
ignored_columns = list(set(tframe.names) - set(x + [y, offset, folds, weights]))
kwargs["ignored_columns"] = None if not ignored_columns else [quoted(col) for col in ignored_columns]
kwargs = dict([(k, kwargs[k].frame_id if isinstance(kwargs[k], H2OFrame) else kwargs[k]) for k in kwargs if
kwargs[k] is not None]) # gruesome one-liner
algo = self.model._compute_algo() # unique to grid search
if self.grid_id is not None: kwargs["grid_id"] = self.grid_id
rest_ver = kwargs.pop("_rest_version") if "_rest_version" in kwargs else None
grid = H2OJob(h2o.api("POST /99/Grid/%s" % algo, data=kwargs), job_type=(algo + " Grid Build"))
if self._future:
self._job = grid
return
grid.poll()
grid_json = h2o.api("GET /99/Grids/%s" % (grid.dest_key))
failure_messages_stacks = ""
error_index = 0
if len(grid_json["failure_details"]) > 0:
print("Errors/Warnings building gridsearch model\n")
# will raise error if no grid model is returned, store error messages here
for error_message in grid_json["failure_details"]:
if isinstance(grid_json["failed_params"][error_index], dict):
for h_name in grid_json['hyper_names']:
print("Hyper-parameter: {0}, {1}".format(h_name,
grid_json['failed_params'][error_index][h_name]))
if len(grid_json["failure_stack_traces"]) > error_index:
print("failure_details: {0}\nfailure_stack_traces: "
"{1}\n".format(error_message, grid_json['failure_stack_traces'][error_index]))
failure_messages_stacks += error_message+'\n'
error_index += 1
self.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]
# get first model returned in list of models from grid search to get model class (binomial, multinomial, etc)
# sometimes no model is returned due to bad parameter values provided by the user.
if len(grid_json['model_ids']) > 0:
first_model_json = h2o.api("GET /%d/Models/%s" %
(rest_ver or 3, grid_json['model_ids'][0]['name']))['models'][0]
self._resolve_grid(grid.dest_key, grid_json, first_model_json)
else:
if len(failure_messages_stacks)>0:
raise ValueError(failure_messages_stacks)
else:
raise ValueError("Gridsearch returns no model due to bad parameter values or other reasons....")
def _model_build(self, x, y, tframe, vframe, kwargs):
kwargs['training_frame'] = tframe
if vframe is not None: kwargs["validation_frame"] = vframe
if is_type(y, int): y = tframe.names[y]
if y is not None: kwargs['response_column'] = y
if not is_type(x, list, tuple): x = [x]
if is_type(x[0], int):
x = [tframe.names[i] for i in x]
offset = kwargs["offset_column"]
folds = kwargs["fold_column"]
weights = kwargs["weights_column"]
ignored_columns = list(set(tframe.names) - set(x + [y, offset, folds, weights]))
kwargs["ignored_columns"] = None if not ignored_columns else [quoted(col) for col in ignored_columns]
kwargs = dict([(k, kwargs[k].frame_id if isinstance(kwargs[k], H2OFrame) else kwargs[k]) for k in kwargs if
kwargs[k] is not None]) # gruesome one-liner
algo = self.model._compute_algo() # unique to grid search
if self.grid_id is not None: kwargs["grid_id"] = self.grid_id
rest_ver = kwargs.pop("_rest_version") if "_rest_version" in kwargs else None
grid = H2OJob(h2o.api("POST /99/Grid/%s" % algo, data=kwargs), job_type=(algo + " Grid Build"))
if self._future:
self._job = grid
return
grid.poll()
grid_json = h2o.api("GET /99/Grids/%s" % (grid.dest_key))
failure_messages_stacks = ""
error_index = 0
if len(grid_json["failure_details"]) > 0:
print("Errors/Warnings building gridsearch model\n")
# will raise error if no grid model is returned, store error messages here
for error_message in grid_json["failure_details"]:
if isinstance(grid_json["failed_params"][error_index], dict):
for h_name in grid_json['hyper_names']:
print("Hyper-parameter: {0}, {1}".format(h_name,
grid_json['failed_params'][error_index][h_name]))
if len(grid_json["failure_stack_traces"]) > error_index:
print("failure_details: {0}\nfailure_stack_traces: "
"{1}\n".format(error_message, grid_json['failure_stack_traces'][error_index]))
failure_messages_stacks += error_message+'\n'
error_index += 1
self.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]
# get first model returned in list of models from grid search to get model class (binomial, multinomial, etc)
# sometimes no model is returned due to bad parameter values provided by the user.
if len(grid_json['model_ids']) > 0:
first_model_json = h2o.api("GET /%d/Models/%s" %
(rest_ver or 3, grid_json['model_ids'][0]['name']))['models'][0]
self._resolve_grid(grid.dest_key, grid_json, first_model_json)
else:
if len(failure_messages_stacks)>0:
raise ValueError(failure_messages_stacks)
else:
raise ValueError("Gridsearch returns no model due to bad parameter values or other reasons....")
def rename_things():
fr = h2o.import_file(tests.locate("smalldata/logreg/prostate.csv"))
fr.frame_id = "mooochooo"
print h2o.ls()
zz = fr[1:2]
zz.show()
zz.frame_id = "black_sheep_LLC"
print h2o.ls()
from h2o.estimators.gbm import H2OGradientBoostingEstimator
m = H2OGradientBoostingEstimator(ntrees=5, max_depth=2)
m.train(x=fr.names[2:], y=fr.names[1], training_frame=fr)
print m.model_id
m.model_id = "my_gbm_model_wwwww"
print h2o.ls()
print h2o.get_model("my_gbm_model_wwwww")
print h2o.ls()
def get_model_by_algo(algo,models_dict):
mod=None
mod_id=None
for m in list(models_dict.keys()):
if m[0:3]==algo:
mod_id=m
mod=h2o.get_model(m)
return mod,mod_id
def stackedensemble(mod):
coef_norm=None
try:
metalearner = h2o.get_model(mod.metalearner()['name'])
coef_norm=metalearner.coef_norm()
except:
pass
return coef_norm
def test_modelselection_gaussian_model_id():
d = h2o.import_file(
path=pyunit_utils.locate("smalldata/logreg/prostate.csv"))
my_y = "GLEASON"
my_x = ["AGE", "RACE", "CAPSULE", "DCAPS", "PSA", "VOL", "DPROS"]
allsubsets_model = modelSelection(seed=12345,
max_predictor_number=7,
mode="allsubsets")
allsubsets_model.train(training_frame=d, x=my_x, y=my_y)
result_frame_allsubsets = allsubsets_model.result()
numRows = result_frame_allsubsets.nrows
modelIDs_allsubsets = allsubsets_model._model_json["output"][
"best_model_ids"]
maxr_model = modelSelection(seed=12345,
max_predictor_number=7,
mode="maxr")
maxr_model.train(training_frame=d, x=my_x, y=my_y)
result_frame_maxr = maxr_model.result()
maxrsweep_model = modelSelection(seed=12345,
max_predictor_number=7,
mode="maxrsweep")
maxrsweep_model.train(training_frame=d, x=my_x, y=my_y)
# make sure results returned by maxr and maxrsweep are the same
pyunit_utils.compare_frames_local(maxr_model.result()[2:4],
maxrsweep_model.result()[2:4],
prob=1.0,
tol=1e-6)
for ind in list(range(numRows)):
model_from_frame_allsubsets = h2o.get_model(
result_frame_allsubsets["model_id"][ind, 0])
pred_frame_allsubsets = model_from_frame_allsubsets.predict(d)
model_from_frame_allsubsets = h2o.get_model(
modelIDs_allsubsets[ind]['name'])
pred_id_allsubsets = model_from_frame_allsubsets.predict(d)
pyunit_utils.compare_frames_local(pred_frame_allsubsets,
pred_id_allsubsets,
prob=1)
model_from_frame_maxr = h2o.get_model(
result_frame_maxr["model_id"][ind, 0])
pred_frame_maxr = model_from_frame_maxr.predict(d)
pyunit_utils.compare_frames_local(pred_frame_allsubsets,
pred_frame_maxr,
prob=1,
tol=1e-6)
def h2o_print_leaderboard(lb_frame, top_n=999999):
df = lb_frame.as_data_frame()
for i in range(0, min(top_n, df.shape[0])):
model_id = df['model_id'][i]
print(df[i:i + 1].to_string(index=False))
best_model = h2o.get_model(model_id)
pprint(h2o_not_default_params_str(best_model))
print()
def test_nfolds_eq_0():
print("Check nfolds = 0 works properly")
ds = import_dataset()
aml = H2OAutoML(project_name="py_aml_nfolds0", nfolds=0, max_models=3, seed=1)
aml.train(y=ds['target'], training_frame=ds['train'])
_, non_se, _ = get_partitioned_model_names(aml.leaderboard)
amodel = h2o.get_model(non_se[0])
assert amodel.params['nfolds']['actual'] == 0
def check_ignore_cols_automl(models,names,x,y):
models = sum(models.as_data_frame().values.tolist(),[])
for model in models:
if "StackedEnsemble" in model:
continue
else:
assert set(h2o.get_model(model).params["ignored_columns"]["actual"]) == set(names) - {y} - set(x), \
"ignored columns are not honored for model " + model
def test_param_enabled():
print("\n=== enabling "+kcvm+" ===")
aml = setup_and_train(True)
models, non_se, se = get_partitioned_model_names(aml.leaderboard)
check_model_property(se, kcvm, False)
check_model_property(non_se, kcvm, True, True, False)
keys = list_keys_in_memory()
tot, cv = len(keys['models']), len(keys['cv_models'])
print("total models in memory = {tot}, among which {cv} CV models".format(tot=tot, cv=cv))
assert tot > 0, "no models left in memory"
expected = len(models) * nfolds
assert cv == expected, "missing CV models in memory, got {actual}, expected {expected}".format(actual=cv, expected=expected)
for m in non_se:
assert h2o.get_model(m).cross_validation_models(), "missing cv models for model "+m
for m in se:
metal = h2o.get_model(h2o.get_model(m).metalearner()['name'])
assert metal.cross_validation_models(), "missing cv models for metalearner of model "+m
def test_param_enabled():
print("\n=== enabling "+kcvm+" ===")
aml = setup_and_train(True)
models, non_se, se = get_partitioned_model_names(aml.leaderboard)
check_model_property(se, kcvm, False)
check_model_property(non_se, kcvm, True, True, True)
keys = list_keys_in_memory()
tot, cv = len(keys['models']), len(keys['cv_models'])
print("total models in memory = {tot}, among which {cv} CV models".format(tot=tot, cv=cv))
assert tot > 0, "no models left in memory"
expected = len(models) * nfolds
assert cv == expected, "missing CV models in memory, got {actual}, expected {expected}".format(actual=cv, expected=expected)
for m in non_se:
assert h2o.get_model(m).cross_validation_models(), "missing cv models for model "+m
for m in se:
metal = h2o.get_model(h2o.get_model(m).metalearner()['name'])
assert metal.cross_validation_models(), "missing cv models for metalearner of model "+m
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 iris_get_model():
iris = h2o.import_file(path=tests.locate("smalldata/iris/iris.csv"))
model = h2o.random_forest(y=iris[4], x=iris[0:4], ntrees=50)
model.show()
model = h2o.get_model(model._id)
model.show()
def iris_get_model():
iris = h2o.import_file(path=pyunit_utils.locate("smalldata/iris/iris.csv"))
model = H2ORandomForestEstimator(ntrees=50)
model.train(y=4, x=list(range(4)), training_frame=iris)
model.show()
model = h2o.get_model(model._id)
model.show()
def cross_validation_models(self): """ Obtain a list of cross-validation models. :return: list of H2OModel objects """ cvmodels = self._model_json["output"]["cross_validation_models"] if cvmodels is None: return None m = [] for p in cvmodels: m.append(h2o.get_model(p["name"])) return m
def iris_get_model(ip,port):
# Connect to h2o
h2o.init(ip,port)
iris = h2o.import_frame(path=h2o.locate("smalldata/iris/iris.csv"))
model = h2o.random_forest(y=iris[4], x=iris[0:4], ntrees=50)
model.show()
model = h2o.get_model(model._id)
model.show()
def _model_build(self, x, y, tframe, vframe, kwargs):
kwargs['training_frame'] = tframe
if vframe is not None: kwargs["validation_frame"] = vframe
if isinstance(y, int): y = tframe.names[y]
if y is not None: kwargs['response_column'] = y
if not isinstance(x, (list,tuple)): x=[x]
if isinstance(x[0], int):
x = [tframe.names[i] for i in x]
offset = kwargs["offset_column"]
folds = kwargs["fold_column"]
weights= kwargs["weights_column"]
ignored_columns = list(set(tframe.names) - set(x + [y,offset,folds,weights]))
kwargs["ignored_columns"] = None if ignored_columns==[] else [h2o.h2o._quoted(col) for col in ignored_columns]
kwargs = dict([(k, kwargs[k].frame_id if isinstance(kwargs[k], H2OFrame) else kwargs[k]) for k in kwargs if kwargs[k] is not None]) # gruesome one-liner
algo = self.model._compute_algo() #unique to grid search
kwargs["_rest_version"] = 99 #unique to grid search
if self.grid_id is not None: kwargs["grid_id"] = self.grid_id
grid = H2OJob(H2OConnection.post_json("Grid/"+algo, **kwargs), job_type=(algo+" Grid Build"))
if self._future:
self._job = grid
return
grid.poll()
if '_rest_version' in list(kwargs.keys()):
grid_json = H2OConnection.get_json("Grids/"+grid.dest_key, _rest_version=kwargs['_rest_version'])
error_index = 0
if len(grid_json["failure_details"]) > 0:
print("Errors/Warnings building gridsearch model\n")
for error_message in grid_json["failure_details"]:
if isinstance(grid_json["failed_params"][error_index], dict):
for h_name in grid_json['hyper_names']:
print("Hyper-parameter: {0}, {1}".format(h_name, grid_json['failed_params'][error_index][h_name]))
if len(grid_json["failure_stack_traces"]) > error_index:
print("failure_details: {0}\nfailure_stack_traces: "
"{1}\n".format(error_message, grid_json['failure_stack_traces'][error_index]))
error_index += 1
else:
grid_json = H2OConnection.get_json("Grids/"+grid.dest_key)
self.models = [h2o.get_model(key['name']) for key in grid_json['model_ids']]
#get first model returned in list of models from grid search to get model class (binomial, multinomial, etc)
# sometimes no model is returned due to bad parameter values provided by the user.
if len(grid_json['model_ids']) > 0:
first_model_json = H2OConnection.get_json("Models/"+grid_json['model_ids'][0]['name'],
_rest_version=kwargs['_rest_version'])['models'][0]
self._resolve_grid(grid.dest_key, grid_json, first_model_json)
else:
raise ValueError("Gridsearch returns no model due to bad parameter values or other reasons....")
def get_modelGBM(ip, port):
prostate = h2o.import_file(path=h2o.locate("smalldata/logreg/prostate.csv"))
prostate.describe()
prostate[1] = prostate[1].asfactor()
prostate_gbm = h2o.gbm(y=prostate[1], x=prostate[2:9], distribution="bernoulli")
prostate_gbm.show()
prostate_gbm.predict(prostate)
model = h2o.get_model(prostate_gbm._id)
model.show()
def get_model_gbm():
prostate = h2o.import_file(path=pyunit_utils.locate("smalldata/logreg/prostate.csv"))
prostate.describe()
prostate[1] = prostate[1].asfactor()
prostate_gbm = H2OGradientBoostingEstimator(distribution="bernoulli")
prostate_gbm.train(x=range(2,9),y=1, training_frame=prostate)
prostate_gbm.show()
prostate_gbm.predict(prostate)
model = h2o.get_model(prostate_gbm.model_id)
model.show()
def check_model_property(model_names, prop_name, present=True, actual_value=None, default_value=None):
for mn in model_names:
model = h2o.get_model(mn)
if present:
assert prop_name in model.params.keys(), \
"missing {prop} in model {model}".format(prop=prop_name, model=mn)
assert actual_value is None or model.params[prop_name]['actual'] == actual_value, \
"actual value for {prop} in model {model} is {val}, expected {exp}".format(prop=prop_name, model=mn, val=model.params[prop_name]['actual'], exp=actual_value)
assert default_value is None or model.params[prop_name]['default'] == default_value, \
"default value for {prop} in model {model} is {val}, expected {exp}".format(prop=prop_name, model=mn, val=model.params[prop_name]['default'], exp=default_value)
else:
assert prop_name not in model.params.keys(), "unexpected {prop} in model {model}".format(prop=prop_name, model=mn)