Forum Discussion
Inefficient function?
- 10 years ago
Hi Aaron,
if that's lightning-fast, let's make match just one row. Therefore you have to create "TeamLookup"-Table like this:
let fnFillDateIntervalls = (Table as table, DateColumn as text, FillUpUntil as date)=> let //DebugParameters: //Table = #"Staff Movements", //DateColumn = "Joined", Source = Table, #"Sorted Rows" = Table.Sort(Source,{{DateColumn, Order.Descending}}), Index = Table.AddIndexColumn(#"Sorted Rows", "Index", 0, 1), Change = Table.TransformColumnTypes(Index,{{DateColumn, Int64.Type}}), #"Added Custom" = Table.AddColumn(Change, "Custom", each [Index]+1), #"Merged Queries" = Table.NestedJoin(#"Added Custom",{"Index"},#"Added Custom",{"Custom"},"NewColumn",JoinKind.LeftOuter), #"Expanded NewColumn" = Table.ExpandTableColumn(#"Merged Queries", "NewColumn", {DateColumn}, {"NewColumn.Date"}), #"Added Custom1" = Table.AddColumn(#"Expanded NewColumn", "Custom.1", each try {Record.Field(_,DateColumn)..[NewColumn.Date]} otherwise {Record.Field(_,DateColumn)..Number.From(FillUpUntil)}), #"Removed Columns" = Table.RemoveColumns(#"Added Custom1",{DateColumn, "Index", "Custom", "NewColumn.Date"}), #"Expanded Custom.1" = Table.ExpandListColumn(#"Removed Columns", "Custom.1"), #"Renamed Columns" = Table.RenameColumns(#"Expanded Custom.1",{{"Custom.1", DateColumn}}), #"Changed Type" = Table.TransformColumnTypes(#"Renamed Columns",{{DateColumn, type date}}) in #"Changed Type", Source = #"Staff Movements", #"Grouped Rows" = Table.Group(Source, {"Name"}, {{"All", each fnFillDateIntervalls(_, "Joined", Date.From(DateTime.LocalNow())), type table}}), #"Expanded All" = Table.ExpandTableColumn(#"Grouped Rows", "All", {"Joined", "Team"}, {"Joined", "Team"}) in #"Expanded All"This query transforms your "Staff Movements"-table into a table with one row for each day. You should disable load to the datamodel. So although this lengthens your table considerably, the result is still very fast, as it used the simple query you mentioned in your last post.
Edit: Don't use Internet-Explorer or Edge to copy this code but preferrably Firefox. Otherwise the code will break (this time will greet you with a request for a comma where there is already one) !!!
There are some things you can try:
1) Group your fact table (Timesheets) on Name and return "All Rows" (Edit: See description here: https://blog.crossjoin.co.uk/2015/05/11/nested-calculations-in-power-query/). This will sort-of partition your big table: Returning only those rows of the table whith that name. You can then omit that condition in the SelectRows-step.
I would think that this will speed up considerably and no further actions necessary.
Other ideas: The sorting is expensive - so you could sort the lookup-table once as a start, pass an index-column to it and buffer that before passing to the function. You then select the MAX index-column instead in your last step.
Also: I've never used Table.IsEmpty - maybe Table.Countrows is faster - but just a vague guess.