Reject unsupported transform methods in getCategoricalMask

getCategoricalMask only models recode/dummycode/hash column expansion.
Specs using bin, word_embedding, bag_of_words, or udf would yield a mask
with the wrong number of columns, so reject them explicitly with a clear
error instead of returning a silently incorrect result. impute and omit
remain accepted since they do not change output arity or the categorical
flag. Add instruction-level tests covering each rejected method and the
accepted impute/omit case.

Author: Baunsgaard

src/main/java/org/apache/sysds/runtime/instructions/cp/BinaryFrameScalarCPInstruction.java

@@ -70,6 +70,19 @@ public void processGetCategorical(ExecutionContext ec, FrameBlock f, ScalarObjec
 				throw new DMLRuntimeException("not supported non ID based spec for get_categorical_mask");
 			}
 
+			// get_categorical_mask only models the column expansion of recode/dummycode/hash.
+			// Methods that change the output arity (bin expands under dummycode, word_embedding and
+			// bag_of_words map to many columns) or are user-defined (udf) would produce a mask with
+			// the wrong number of columns, so reject them explicitly instead of emitting a silently
+			// incorrect result. impute and omit are intentionally allowed: they do not alter the
+			// output column count or the categorical flag of a column.
+			for(TfMethod m : new TfMethod[] {TfMethod.BIN, TfMethod.WORD_EMBEDDING, TfMethod.BAG_OF_WORDS,
+				TfMethod.UDF}) {
+				if(jSpec.containsKey(m.toString()))
+					throw new DMLRuntimeException(
+						"unsupported transform method '" + m + "' for get_categorical_mask");
+			}
+
 			String recode = TfMethod.RECODE.toString();
 			String dummycode = TfMethod.DUMMYCODE.toString();
 			String hash = TfMethod.HASH.toString();

src/test/java/org/apache/sysds/test/component/frame/transform/GetCategoricalMaskInstructionTest.java

@@ -98,6 +98,50 @@ public void nonIdSpecIdsFalseThrows() {
 		assertThrowsMessage("non ID based spec", () -> run(meta, "{\"ids\": false, \"recode\": [1]}"));
 	}
 
+	@Test
+	public void unsupportedBinMethodThrows() {
+		// bin expands to bin-count columns under dummycode, which the mask does not model
+		FrameBlock meta = new FrameBlock(new ValueType[] {ValueType.STRING}, new String[][] {{"a"}});
+		assertThrowsMessage("unsupported transform method 'bin'",
+			() -> run(meta, "{\"ids\": true, \"bin\": [{\"id\": 1, \"method\": \"equi-width\", \"numbins\": 3}]}"));
+	}
+
+	@Test
+	public void unsupportedWordEmbeddingMethodThrows() {
+		// word_embedding maps a column to an embedding vector (many columns), not a single mask entry
+		FrameBlock meta = new FrameBlock(new ValueType[] {ValueType.STRING}, new String[][] {{"a"}});
+		assertThrowsMessage("unsupported transform method 'word_embedding'",
+			() -> run(meta, "{\"ids\": true, \"word_embedding\": [1]}"));
+	}
+
+	@Test
+	public void unsupportedBagOfWordsMethodThrows() {
+		// bag_of_words expands to one column per dictionary token
+		FrameBlock meta = new FrameBlock(new ValueType[] {ValueType.STRING}, new String[][] {{"a"}});
+		assertThrowsMessage("unsupported transform method 'bag_of_words'",
+			() -> run(meta, "{\"ids\": true, \"bag_of_words\": [1]}"));
+	}
+
+	@Test
+	public void unsupportedUdfMethodThrows() {
+		// udf output arity is user-defined and cannot be inferred from the spec
+		FrameBlock meta = new FrameBlock(new ValueType[] {ValueType.STRING}, new String[][] {{"a"}});
+		assertThrowsMessage("unsupported transform method 'udf'",
+			() -> run(meta, "{\"ids\": true, \"udf\": {\"name\": \"f\", \"ids\": [1]}}"));
+	}
+
+	@Test
+	public void imputeAndOmitAreAccepted() {
+		// impute and omit do not change the output column count or categorical flag, so a spec that
+		// only adds them on top of a recoded column must still succeed and mark that column categorical
+		FrameBlock meta = new FrameBlock(new ValueType[] {ValueType.STRING}, new String[][] {{"a"}});
+		MatrixBlock res = run(meta, "{\"ids\": true, \"recode\": [1], \"impute\": [{\"id\": 1, \"method\": \"global_mode\"}], \"omit\": [1]}");
+
+		assertEquals(1, res.getNumRows());
+		assertEquals(1, res.getNumColumns());
+		assertEquals(1.0, res.get(0, 0), 0.0);
+	}
+
 	@Test
 	public void unsupportedOpcodeThrows() {
 		// any frame-scalar binary opcode other than get_categorical_mask must be rejected