feat: Add Hyperparameter the GetGroups call with aggregate values

master/internal/api_runs.go

@@ -1091,9 +1091,8 @@ func (a *apiServer) GetRunGroups(ctx context.Context, req *apiv1.GetRunGroupsReq
 
 	resp := &apiv1.GetRunGroupsResponse{}
 	var groups []*runv1.RunGroup
-	query := db.Bun().NewSelect().
-		Model(&groups).
-		ModelTableExpr("runs AS r").
+	searchQuery := db.Bun().NewSelect().
+		TableExpr("runs AS r").
 		Apply(getRunsGroupsColumns)
 
 	var proj *projectv1.Project
@@ -1103,36 +1102,64 @@ func (a *apiServer) GetRunGroups(ctx context.Context, req *apiv1.GetRunGroupsReq
 			return nil, err
 		}
 
-		query = query.Where("r.project_id = ?", req.ProjectId)
+		searchQuery = searchQuery.Where("r.project_id = ?", req.ProjectId)
 	}
-	if query, err = experiment.AuthZProvider.Get().
-		FilterExperimentsQuery(ctx, *curUser, proj, query,
+	if searchQuery, err = experiment.AuthZProvider.Get().
+		FilterExperimentsQuery(ctx, *curUser, proj, searchQuery,
 			[]rbacv1.PermissionType{rbacv1.PermissionType_PERMISSION_TYPE_VIEW_EXPERIMENT_METADATA},
 		); err != nil {
 		return nil, err
 	}
 	if req.Filter != nil {
-		query, err = filterRunQuery(query, req.Filter)
+		searchQuery, err = filterRunQuery(searchQuery, req.Filter)
 		if err != nil {
 			return nil, err
 		}
 	}
 
 	if req.Sort != nil {
-		err = sortRuns(req.Sort, query, false)
+		err = sortRuns(req.Sort, searchQuery, false)
 		if err != nil {
 			return nil, err
 		}
 	} else {
-		query.OrderExpr("group_name ASC")
+		searchQuery.OrderExpr("group_name ASC")
 	}
 
 	groupName, err := runColumnNameToSQL(req.Group)
 	if err != nil {
 		return nil, err
 	}
-	query.Group(groupName)
-	query.ColumnExpr(fmt.Sprintf("%s AS group_name", groupName))
+	searchQuery.Group(groupName)
+	searchQuery.ColumnExpr(fmt.Sprintf("%s AS group_name", groupName))
+
+	query := db.Bun().NewSelect().
+		ColumnExpr("run_groups.group_name").
+		ColumnExpr("run_groups.start_time").
+		ColumnExpr("run_groups.end_time").
+		ColumnExpr("run_groups.checkpoint_size").
+		ColumnExpr("run_groups.checkpoint_count").
+		ColumnExpr("run_groups.searcher_metric_value").
+		ColumnExpr("run_groups.duration").
+		ColumnExpr("run_groups.user_ids").
+		ColumnExpr("run_groups.resource_pools").
+		ColumnExpr("run_groups.searcher_types").
+		ColumnExpr("run_groups.searcher_metrics").
+		ColumnExpr("run_groups.run_count").
+		ColumnExpr("array_to_json(run_groups.run_ids) as run_ids").
+		ColumnExpr(`(SELECT jsonb_concat_agg(jsonb_build_object(s.hparam,
+			jsonb_build_object(
+				'number_val', s.number_val,
+				'text_val', s.text_val,
+				'bool_val', s.bool_val)
+				)) FROM (
+					SELECT hparam, avg(number_val) as number_val,
+						json_agg(distinct coalesce(text_val, '')) as text_val,
+						bool_or(bool_val) as bool_val
+					FROM run_hparams rh WHERE rh.run_id = ANY (run_groups.run_ids)
+					GROUP BY rh.hparam) as s) as hyperparameters`).
+		Model(&groups).
+		ModelTableExpr("(?) as run_groups", searchQuery)
 	pagination, err := runPagedBunExperimentsQuery(ctx, query, int(req.Offset), int(req.Limit))
 	if err != nil {
 		return nil, err
@@ -1155,6 +1182,7 @@ func getRunsGroupsColumns(q *bun.SelectQuery) *bun.SelectQuery {
 		ColumnExpr("json_agg(distinct e.config->'searcher'->>'name') as searcher_types").
 		ColumnExpr("json_agg(distinct e.config->'searcher'->>'metric') as searcher_metrics").
 		ColumnExpr("COUNT(*) AS run_count").
+		ColumnExpr("array_agg(r.id) as run_ids").
 		Join("LEFT JOIN experiments AS e ON r.experiment_id=e.id").
 		Join("LEFT JOIN runs_metadata AS rm ON r.id=rm.run_id").
 		Join("LEFT JOIN users u ON e.owner_id = u.id").

master/internal/api_runs_intg_test.go

@@ -1790,7 +1790,12 @@ func TestGetRunGroups(t *testing.T) {
 		Sort:      ptrs.Ptr("state=asc"),
 	}
 
-	hyperparameters := map[string]any{"global_batch_size": 1, "test1": map[string]any{"test2": 1}}
+	hyperparameters := map[string]any{
+		"global_batch_size": 1,
+		"test1":             map[string]any{"test2": 1},
+		"stringH":           "abc",
+		"boolH":             false,
+	}
 
 	exp := createTestExpWithProjectID(t, api, curUser, projectIDInt)
 
@@ -1839,4 +1844,45 @@ func TestGetRunGroups(t *testing.T) {
 	resp, err = api.GetRunGroups(ctx, req)
 	require.NoError(t, err)
 	require.Len(t, resp.Groups, 1)
+
+	// Add new task with different hyperperameter values
+	newHparams := map[string]any{
+		"global_batch_size": 9,
+		"test1":             map[string]any{"test2": 8},
+		"stringH":           "def",
+		"boolH":             true,
+	}
+
+	exp3 := createTestExpWithProjectID(t, api, curUser, projectIDInt)
+
+	task = &model.Task{TaskType: model.TaskTypeTrial, TaskID: model.NewTaskID()}
+	require.NoError(t, db.AddTask(ctx, task))
+	require.NoError(t, db.AddTrial(ctx, &model.Trial{
+		State:        model.PausedState,
+		ExperimentID: exp3.ID,
+		StartTime:    time.Now(),
+		HParams:      newHparams,
+	}, task.TaskID))
+
+	req = &apiv1.GetRunGroupsRequest{
+		ProjectId: &projectID,
+		Group:     "state",
+		Sort:      ptrs.Ptr("state=asc"),
+	}
+
+	resp, err = api.GetRunGroups(ctx, req)
+	require.NoError(t, err)
+	require.Len(t, resp.Groups, 2)
+	require.Equal(t, string(model.PausedState), resp.Groups[0].GroupName)
+	require.Equal(t, string(model.CanceledState), resp.Groups[1].GroupName)
+	require.InEpsilon(t, float64(5), resp.Groups[0].Hyperparameters.Fields["global_batch_size"].
+		GetStructValue().Fields["number_val"].GetNumberValue(), 0.00001)
+	require.InEpsilon(t, 4.5, resp.Groups[0].Hyperparameters.Fields["test1.test2"].
+		GetStructValue().Fields["number_val"].GetNumberValue(), 0.00001)
+	require.Equal(t, "abc", resp.Groups[0].Hyperparameters.Fields["stringH"].
+		GetStructValue().Fields["text_val"].GetListValue().Values[0].GetStringValue())
+	require.Equal(t, "def", resp.Groups[0].Hyperparameters.Fields["stringH"].
+		GetStructValue().Fields["text_val"].GetListValue().Values[1].GetStringValue())
+	require.True(t, resp.Groups[0].Hyperparameters.Fields["boolH"].
+		GetStructValue().Fields["bool_val"].GetBoolValue())
 }

master/static/migrations/20241031154311_add-json-concat-aggregate.tx.up.sql

@@ -0,0 +1,4 @@
+create aggregate jsonb_concat_agg(jsonb)(
+	sfunc = jsonb_concat(jsonb, jsonb),
+	stype = jsonb
+);

performance/daist/daist/rest_api/tasks.py

@@ -92,6 +92,10 @@ def read_only_tasks(resources: Resources) -> LocustTasksWithMeta:
             tasks.append(LocustGetTaskWithMeta(
                 f"/api/v1/projects/{resources.project_id}/columns",
                 test_name="get project columns"))
+
+        tasks.append(LocustGetTaskWithMeta(
+            f"/api/v1/runs/groups?projectId={resources.project_id}&group=state",
+            test_name="get groups"))
 
     if resources.workspace_id is not None:
         tasks.append(LocustGetTaskWithMeta(

proto/src/determined/run/v1/run.proto

@@ -268,4 +268,8 @@ message RunGroup {
   repeated string resource_pools = 11;
   // The number of runs in the group.
   int32 run_count = 12;
+  // The run ids of the runs that belong to the group
+  repeated int32 run_ids = 13;
+  // hyperparameters.
+  optional google.protobuf.Struct hyperparameters = 14;
 }