DRF can create leaf nodes that train observations never visit due to NAs

[exalate-issue-sync]

Repro on 3.32.1.2 example:

{code:python}import h2o
from h2o.estimators import H2ORandomForestEstimator
h2o.init()

h2o_df = h2o.import_file("/Users/jgranados/datasets/BNPParibas.csv")

predictors = h2o_df.columns
response = "v58"
predictors.remove(response)

cars_drf = H2ORandomForestEstimator(
ntrees=1,
max_depth=10,
min_rows=1,
nbins=4096,
nbins_top_level=4096,
min_split_improvement=1e-06,
seed=3
)
cars_drf.train(x=predictors,
y=response,
training_frame=h2o_df,
)

from h2o.tree import H2OTree

node_assignments_num = cars_drf.predict_leaf_node_assignment(h2o_df, type="Node_ID")

tree = H2OTree(model = cars_drf, tree_number=0)
children = []
for i in range(0, len(tree)):
if (tree.left_children[i] == -1):
children.append(tree.node_ids[i])

print(len(node_assignments_num.unique()))
print(len(children))
len(node_assignments_num.unique()) == len(children){code}

Output:

{noformat}Parse progress: |█████████████████████████████████████████████████████████| 100%
drf Model Build progress: |███████████████████████████████████████████████| 100%
829
830

False{noformat}

[h2o-ops]

JIRA Issue Details

Jira Issue: PUBDEV-8242
Assignee: Michal Kurka
Reporter: Joseph Granados
State: Resolved
Fix Version: 3.32.1.5
Attachments: N/A
Development PRs: Available

[h2o-ops]

Linked PRs from JIRA

#5577
#5897