def import_dataset():
df = h2o.import_file(path=pu.locate("smalldata/prostate/prostate.csv"))
target = "CAPSULE"
target_alt = "RACE"
df[target] = df[target].asfactor()
df[target_alt] = df[target_alt].asfactor()
return pu.ns(train=df, target=target, target_alt=target_alt)
def prepare_data():
fr = h2o.import_file(path=pu.locate("smalldata/junit/weather.csv"))
target = "RainTomorrow"
fr[target] = fr[target].asfactor()
ds = pu.ns(x=fr.columns, y=target, train=fr)
return ds
def compute_perf(model):
perf = pu.ns(train=model.model_performance(train=True),
test=model.model_performance(test_data=ds.test))
print("{} training performance: ".format(model.model_id))
print(perf.train)
print("{} test performance: ".format(model.model_id))
print(perf.test)
return perf
def import_dataset(seed=0, mode='binary'):
df = h2o.import_file(
path=pu.locate("smalldata/titanic/titanic_expanded.csv"), header=1)
target = dict(binary='survived', multiclass='pclass',
regression='fare')[mode]
fr = df.split_frame(ratios=[.8], seed=seed)
return pu.ns(train=fr[0], test=fr[1], target=target)
def import_dataset(seed=0, larger=False):
df = h2o.import_file(path=pu.locate("smalldata/prostate/{}".format("prostate_complete.csv.zip" if larger else "prostate.csv")))
target = "CAPSULE"
df[target] = df[target].asfactor()
#Split frames
fr = df.split_frame(ratios=[.8,.1], seed=seed)
#Set up train, validation, and test sets
return pu.ns(train=fr[0], valid=fr[1], test=fr[2], target=target, target_idx=1)
def compute_perf(model):
perf = pu.ns(
train=model.model_performance(train=True),
test=model.model_performance(test_data=ds.test)
)
print("{} training performance: ".format(model.model_id))
print(perf.train)
print("{} test performance: ".format(model.model_id))
print(perf.test)
return perf
def import_dataset(seed=0):
df = h2o.import_file(path=pu.locate("smalldata/prostate/prostate.csv"))
target = "CAPSULE"
df[target] = df[target].asfactor()
fr = df.split_frame(ratios=[.8, .1], seed=seed)
return pu.ns(train=fr[0],
valid=fr[1],
test=fr[2],
target=target,
target_idx=1)
def prepare_data(blending=False):
fr = h2o.import_file(path=pu.locate("smalldata/testng/higgs_train_5k.csv"))
target = "response"
fr[target] = fr[target].asfactor()
ds = pu.ns(x=fr.columns, y=target, train=fr)
if blending:
train, blend = fr.split_frame(ratios=[.7], seed=seed)
return ds.extend(train=train, blend=blend)
else:
return ds
def prepare_data(blending=False):
fr = h2o.import_file(path=pu.locate("smalldata/junit/weather.csv"))
target = "RainTomorrow"
fr[target] = fr[target].asfactor()
ds = pu.ns(x=fr.columns, y=target, train=fr)
if blending:
train, blend = fr.split_frame(ratios=[.7], seed=seed)
return ds.extend(train=train, blend=blend)
else:
return ds
def prepare_data(blending=False):
fr = h2o.import_file(path=pu.locate("smalldata/testng/higgs_train_5k.csv"))
target = "response"
fr[target] = fr[target].asfactor()
ds = pu.ns(x=fr.columns, y=target, train=fr)
if blending:
train, blend = fr.split_frame(ratios=[.7], seed=seed)
return ds.extend(train=train, blend=blend)
else:
return ds
def load_dataset(incl_test=False, incl_foldc=False):
fr = h2o.import_file(pu.locate("smalldata/titanic/titanic_expanded.csv"), header=1)
target = "pclass"
train = fr
test = None
if incl_test:
fr = fr.split_frame(ratios=[.8], destination_frames=["titanic_train", "titanic_test"], seed=seed)
train = fr[0]
test = fr[1]
if incl_foldc:
train["foldc"] = train.kfold_column(3, seed)
return pu.ns(train=train, test=test, target=target)
def test_base_models_can_use_different_compatible_training_frames():
"""
test that passing in base models that use different subsets of
the features works. (different training_frame)
"""
ds = prepare_data(blending)
datasets = pu.ns(gbm=ds.extend(x=None,
train=ds.train[list(range(1, 11))].cbind(ds.train[ds.y])),
drf=ds.extend(x=None,
train=ds.train[list(range(13, 20))].cbind(ds.train[ds.y])))
bm = train_base_models(datasets)
se = train_stacked_ensemble(ds, bm)
assert se.auc() > 0
def test_se_fails_when_base_models_use_incompatible_training_frames():
"""
test that SE fails when passing in base models that were trained with frames of different size
"""
ds = prepare_data(blending)
datasets = pu.ns(gbm=ds.extend(x=None),
drf=ds.extend(x=None, train=ds.train[0:ds.train.nrows//2,:]))
bm = train_base_models(datasets)
try:
train_stacked_ensemble(ds, bm)
assert False, "Stacked Ensembles of models with different training frame sizes should fail"
except Exception as e:
assert "Base models are inconsistent: they use different size (number of rows) training frames" in str(e), "wrong error message: {}".format(str(e))
def test_base_models_can_use_different_x():
"""
test that passing in base models that use different subsets of
the features works. (different x, but same training_frame)
"""
ds = prepare_data(blending)
datasets = pu.ns(gbm=ds.extend(x=ds.x[1:11]),
drf=ds.extend(x=ds.x[13:20]))
bm = train_base_models(datasets)
se = train_stacked_ensemble(ds, bm)
se_nox = train_stacked_ensemble(ds.extend(x=None), bm)
assert se.auc() > 0
assert se.auc() == se_nox.auc()
def test_base_models_can_use_different_x():
"""
test that passing in base models that use different subsets of
the features works. (different x, but same training_frame)
"""
ds = prepare_data(blending)
datasets = pu.ns(gbm=ds.extend(x=ds.x[1:11]),
drf=ds.extend(x=ds.x[13:20]))
bm = train_base_models(datasets)
se = train_stacked_ensemble(ds, bm)
se_nox = train_stacked_ensemble(ds.extend(x=None), bm)
assert se.auc() > 0
assert se.auc() == se_nox.auc()
def setup_data():
# MNIST is multinomial classification problem
train_full = h2o.import_file(
pu.locate("bigdata/laptop/mnist/train.csv.gz"))
# test_full = h2o.import_file(pu.locate("bigdata/laptop/mnist/test.csv.gz"))
# train = train_full
# test = test_full
train, test, _ = train_full.split_frame(ratios=[.05, .1], seed=seed)
x = train.columns[:-1]
y = -1
for fr in [train]:
fr[y] = fr[y].asfactor()
domain = unique(train[y])
print(domain)
return pu.ns(x=x, y=y, train=train, test=test, domain=domain)
def prepare_data(blending=False):
col_types = ["numeric", "numeric", "numeric", "enum", "enum", "numeric", "numeric", "numeric", "numeric"]
dat = h2o.upload_file(path=pu.locate("smalldata/extdata/prostate.csv"),
destination_frame="prostate_hex",
col_types=col_types)
train, test = dat.split_frame(ratios=[.8], seed=1)
x = ["CAPSULE", "GLEASON", "RACE", "DPROS", "DCAPS", "PSA", "VOL"]
y = "AGE"
ds = pu.ns(x=x, y=y, train=train, test=test)
if blending:
train, blend = train.split_frame(ratios=[.7], seed=seed)
return ds.extend(train=train, blend=blend)
else:
return ds
def test_base_models_can_use_different_compatible_training_frames():
"""
test that passing in base models that use different subsets of
the features works. (different training_frame)
"""
ds = prepare_data(blending)
datasets = pu.ns(gbm=ds.extend(x=None,
train=ds.train[list(range(
1, 11))].cbind(ds.train[ds.y])),
drf=ds.extend(x=None,
train=ds.train[list(range(
13, 20))].cbind(ds.train[ds.y])))
bm = train_base_models(datasets)
se = train_stacked_ensemble(ds, bm)
assert se.auc() > 0
def test_se_fails_when_base_models_use_incompatible_training_frames():
"""
test that SE fails when passing in base models that were trained with frames of different size
"""
ds = prepare_data(blending)
datasets = pu.ns(gbm=ds.extend(x=None),
drf=ds.extend(x=None,
train=ds.train[0:ds.train.nrows //
2, :]))
bm = train_base_models(datasets)
try:
train_stacked_ensemble(ds, bm)
assert False, "Stacked Ensembles of models with different training frame sizes should fail"
except Exception as e:
assert "Base models are inconsistent: they use different size (number of rows) training frames" in str(
e), "wrong error message: {}".format(str(e))
def prepare_data(blending=False):
col_types = [
"numeric", "numeric", "numeric", "enum", "enum", "numeric", "numeric",
"numeric", "numeric"
]
dat = h2o.upload_file(path=pu.locate("smalldata/extdata/prostate.csv"),
destination_frame="prostate_hex",
col_types=col_types)
train, test = dat.split_frame(ratios=[.8], seed=1)
x = ["CAPSULE", "GLEASON", "RACE", "DPROS", "DCAPS", "PSA", "VOL"]
y = "AGE"
ds = pu.ns(x=x, y=y, train=train, test=test)
if blending:
train, blend = train.split_frame(ratios=[.7], seed=seed)
return ds.extend(train=train, blend=blend)
else:
return ds
def scores_and_preds(models, test):
retval = lambda d: d
if not isinstance(models, list):
models = [models]
retval = lambda d: next(iter(d.values()))
training_scores = {m.key: m.mean_per_class_error() for m in models}
cv_scores = {m.key: m.mean_per_class_error(xval=True) for m in models}
test_scores = {
m.key: m.model_performance(test).mean_per_class_error()
for m in models
}
test_predictions = {m.key: m.predict(test) for m in models}
test_pclasses = {
m.key: unique(test_predictions[m.key]['predict'])
for m in models
}
return pu.ns(
training_scores=retval(training_scores),
cv_scores=retval(cv_scores),
test_scores=retval(test_scores),
test_pclasses=retval(test_pclasses),
)
def import_dataset():
df = h2o.import_file(path=pu.locate("smalldata/extdata/australia.csv"))
fr = df.split_frame(ratios=[.8, .1])
target = "runoffnew"
return pu.ns(target=target, train=fr[0], valid=fr[1], test=fr[2])
