davidradl commented on code in PR #26667: URL: https://github.com/apache/flink/pull/26667#discussion_r2177605449
########## flink-table/flink-table-planner/src/main/java/org/apache/flink/table/planner/plan/rules/logical/ExpandMLEvaluateTableFunctionRule.java: ########## @@ -0,0 +1,262 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.table.planner.plan.rules.logical; + +import org.apache.flink.annotation.Internal; +import org.apache.flink.table.api.ValidationException; +import org.apache.flink.table.catalog.ContextResolvedFunction; +import org.apache.flink.table.functions.FunctionIdentifier; +import org.apache.flink.table.ml.AsyncPredictRuntimeProvider; +import org.apache.flink.table.ml.PredictRuntimeProvider; +import org.apache.flink.table.ml.TaskType; +import org.apache.flink.table.planner.calcite.FlinkContext; +import org.apache.flink.table.planner.calcite.FlinkTypeFactory; +import org.apache.flink.table.planner.calcite.RexModelCall; +import org.apache.flink.table.planner.functions.bridging.BridgingSqlAggFunction; +import org.apache.flink.table.planner.functions.sql.ml.MLEvaluationAggregationFunction; +import org.apache.flink.table.planner.functions.sql.ml.SqlMLEvaluateTableFunction; +import org.apache.flink.table.planner.functions.sql.ml.SqlMLPredictTableFunction; +import org.apache.flink.table.planner.plan.nodes.logical.FlinkLogicalTableFunctionScan; +import org.apache.flink.table.planner.utils.ShortcutUtils; + +import org.apache.calcite.plan.RelOptRule; +import org.apache.calcite.plan.RelOptRuleCall; +import org.apache.calcite.plan.RelRule; +import org.apache.calcite.rel.RelCollations; +import org.apache.calcite.rel.core.AggregateCall; +import org.apache.calcite.rel.logical.LogicalTableFunctionScan; +import org.apache.calcite.rel.type.RelDataType; +import org.apache.calcite.rex.RexCall; +import org.apache.calcite.rex.RexCallBinding; +import org.apache.calcite.rex.RexLiteral; +import org.apache.calcite.rex.RexNode; +import org.apache.calcite.sql.SqlKind; +import org.apache.calcite.tools.RelBuilder; +import org.apache.calcite.util.NlsString; +import org.immutables.value.Value; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.List; +import java.util.stream.Collectors; + +/** + * Rule that expands ML evaluation table function calls. + * + * <p>This rule matches {@link FlinkLogicalTableFunctionScan} with a {@link + * SqlMLEvaluateTableFunction} call and expands it into ml predict table function and an aggregation + * function following it. + */ +@Internal [email protected] +public class ExpandMLEvaluateTableFunctionRule + extends RelRule<ExpandMLEvaluateTableFunctionRule.Config> { + + public static final RelOptRule INSTANCE = new ExpandMLEvaluateTableFunctionRule(Config.DEFAULT); + + public ExpandMLEvaluateTableFunctionRule(Config config) { + super(config); + } + + @Override + public void onMatch(RelOptRuleCall call) { + final LogicalTableFunctionScan scan = call.rel(0); + final RelDataType resultType = scan.getRowType(); + final RelBuilder relBuilder = call.builder().push(scan.getInput(0)); + + final RexCall rexCall = (RexCall) scan.getCall(); + + RelDataType predictOutputType = addPredictTableFunction(relBuilder, rexCall); + addProjection(relBuilder, rexCall, predictOutputType); + addAggregate(relBuilder, rexCall, resultType); + + call.transformTo(relBuilder.build()); + } + + private void addAggregate(RelBuilder relBuilder, RexCall rexCall, RelDataType resultType) { + final String task = getTask(rexCall); + final MLEvaluationAggregationFunction aggregationFunction = + new MLEvaluationAggregationFunction(task); + final FlinkContext context = ShortcutUtils.unwrapContext(relBuilder.getCluster()); + final FlinkTypeFactory typeFactory = + ShortcutUtils.unwrapTypeFactory(relBuilder.getCluster()); + relBuilder.aggregate( + relBuilder.groupKey(), + List.of( + AggregateCall.create( + BridgingSqlAggFunction.of( + context, + typeFactory, + ContextResolvedFunction.temporary( + FunctionIdentifier.of("ml_evaluate"), + aggregationFunction)), + false, + false, + false, + List.of(0, 1), + -1, + null, + RelCollations.EMPTY, + resultType.getFieldList().get(0).getType(), + "result"))); + } + + private void addProjection( + RelBuilder relBuilder, RexCall rexCall, RelDataType predictOutputType) { + final RexCall labelDescriptor = (RexCall) rexCall.getOperands().get(2); + final String labelCol = + ((RexLiteral) labelDescriptor.getOperands().get(0)) + .getValueAs(NlsString.class) + .getValue(); + + // Project the label column and the last column (prediction). Only one label and prediction + // column is expected. Validation is done in SqlMLEvaluateTableFunction. + final List<RexNode> projectFields = + predictOutputType.getFieldList().stream() + .filter( + field -> + field.getName().equals(labelCol) + || field.getIndex() + == predictOutputType.getFieldCount() - 1) + .map(field -> relBuilder.field(field.getIndex())) + .collect(Collectors.toList()); + relBuilder.project(projectFields); + } + + private RelDataType addPredictTableFunction(RelBuilder relBuilder, RexCall rexCall) { + final RexCall tableArg = (RexCall) rexCall.getOperands().get(0); + final RexCall modelCall = (RexCall) rexCall.getOperands().get(1); + final RexCall featuresDescriptor = (RexCall) rexCall.getOperands().get(3); + + // Get optional config map if present + final RexCall configMap = getConfigMap(rexCall); + + final List<RexNode> predictOperands = new ArrayList<>(); + predictOperands.add(tableArg); + predictOperands.add(modelCall); + predictOperands.add(featuresDescriptor); + if (configMap != null) { + predictOperands.add(configMap); + } + + final RexCall predictCall = + (RexCall) + relBuilder + .getRexBuilder() + .makeCall(new SqlMLPredictTableFunction(), predictOperands); + + RexCallBinding callBinding = + new RexCallBinding( + relBuilder.getTypeFactory(), + predictCall.getOperator(), + predictOperands, + List.of()); + + RelDataType predictReturnType = + ((SqlMLPredictTableFunction) predictCall.getOperator()) + .getRowTypeInference() + .inferReturnType(callBinding); + + relBuilder.push( + LogicalTableFunctionScan.create( + relBuilder.getCluster(), + List.of(relBuilder.build()), + predictCall, + null, + predictReturnType, + Collections.emptySet())); + + return predictReturnType; + } + + private static RexCall getConfigMap(RexCall rexCall) { + if (rexCall.getOperands().size() > 5) { + return (RexCall) rexCall.getOperands().get(5); + } + if (rexCall.getOperands().size() > 4) { + RexNode node = rexCall.getOperands().get(4); + if (node instanceof RexCall + && ((RexCall) node).getOperator().getKind() == SqlKind.MAP_VALUE_CONSTRUCTOR) { + return (RexCall) node; + } + } + return null; + } + + private static String getTask(RexCall rexCall) { + final RexNode taskNode = rexCall.getOperands().get(4); + String task = null; + if (taskNode instanceof RexLiteral) { + task = ((RexLiteral) taskNode).getValueAs(NlsString.class).getValue(); + if (task == null || task.isEmpty()) { Review Comment: we do not need to `if (task == null` if we are going to set it to null -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
