Skip to content

GH-16719: Target encoding: Add feature interaction with known domain test is sometimes failing #16729

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
tomasfryda wants to merge 2 commits into
base: rel-3.46.0
Choose a base branch
from
Open

GH-16719: Target encoding: Add feature interaction with known domain test is sometimes failing #16729

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
aa0e646
Fix NewChunk's behavior regarding categorical cancel_sparse
tomasfryda Jan 8, 2026
7da8fca
Formatting and check data consistency in the test
tomasfryda Jan 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 23 additions & 5 deletions h2o-core/src/main/java/water/fvec/NewChunk.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -517,16 +517,34 @@ protected final boolean isCategorical(int idx) {
}

public void addCategorical(int e) {
if(_ms == null || _ms.len() == _sparseLen)
append2slow();
if( e != 0 || !isSparseZero() ) {
_ms.set(_sparseLen,e);
if (_ms == null || _ms.len() == _sparseLen) {
if (2 * _sparseLen > _len && isSparseZero()) {
int ids[] = new int[_id.length + 1];
System.arraycopy(_id, 0, ids, 1, _id.length);
ids[0] = -1;
// append2slow() will run cancel_sparse which will lose the information about 0 being a category.
// To be safe just set to categorical only those zeros that are surrounded by categories.
// We care only about those zeros set before cancel_sparse was called after the call all zeros
// should be set correctly as category since isSparseZero() == false.
append2slow();
for (int i = 0; i < ids.length - 1; i++) {
if (_xs.isCategorical(ids[i + 1]) && (i == 0 || _xs.isCategorical(ids[i])))
for (int j = ids[i] + 1; j < ids[i + 1]; j++) {
_xs.setCategorical(j);
}
}
} else
append2slow();
Comment thread
tomasfryda marked this conversation as resolved.
}
if (e != 0 || !isSparseZero()) {
_ms.set(_sparseLen, e);
_xs.setCategorical(_sparseLen);
if(_id != null) _id[_sparseLen] = _len;
if (_id != null) _id[_sparseLen] = _len;
++_sparseLen;
}
++_len;
}

public void addNA() {
if(!_sparseNA) {
if (isString()) {
Expand Down
30 changes: 30 additions & 0 deletions h2o-core/src/test/java/water/fvec/NewChunkTest.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -629,5 +629,35 @@ public void testSparseWithMissing() {
nc.addNumDecompose(Double.MIN_VALUE);
nc.addNumDecompose(Double.MIN_NORMAL);
}

@Test
public void testNewChunkSparseCategoricals() {
// this is reproduction of issue that caused failure in multinode for the
// targetencoding.interaction.InteractionSupportTest.test_addFeatureInteraction_with_known_domain_is_consistent
// test
int[] data = new int[96];
for (int i = 0; i < 19; i++)
data[i] = 0;
for (int i = 19; i < 62; i++)
data[i] = 2;
for (int i = 62; i < 86; i++)
data[i] = 7;
for (int i = 86; i < 90; i++)
data[i] = 12;
for (int i = 90; i < 94; i++)
data[i] = 0;
data[94] = 11;
data[95] = 11;

nc = new NewChunk(av, 25);
for (int i = 0; i < data.length; i++) {
nc.addCategorical(data[i]);
}
assertEquals(Vec.T_CAT, nc.type());
Chunk c = nc.new_close();
for (int i = 0; i < data.length; i++) {
assertEquals(data[i], (int)c.atd(i));
}
}
}

Loading