Hi Julian
yes, the data is all on screen, I've also attached a spreadsheet with the data so you can play with it yourself.
The product is a Winter season only product, the Prior Year data PY shows sales in the Winter months, but no sales in the Summer then sales increasing again towards the next Winter, then decreasing again towards the Spring, we then want to forecast the remainder of the current year and sales for next year too.
The forecasting algorithms don't seem to be recognising the pattern of historic data, just the decreasing sales in the Spring of the current year and assuming the sales will continue that trend below zero into negative sales.
We need to understand if this is a limitation of the forecasting functionality and if so then not use it, or use it with caution, think it is especially important to have an option to be able to stop forecast values going negative, especially of being used for sales forecasting.
Regards
Mark
------------------------------
Mark White
------------------------------
Original Message:
Sent: Tue October 13, 2020 10:49 AM
From: Julian
Subject: PAW 57 Forecasting Time Dimension Detection
Hi Mark, is all the historical data displayed on the screenshot above? given the trend on the screenshot, in principle, the forecast looks accurate (forecasting do not apply any constraints i.e "sales cannot be negative"). The high and low lines are to indicate the confidence interval.
Julian
------------------------------
Julian Solano Acosta
Original Message:
Sent: Mon October 12, 2020 12:08 PM
From: Mark White
Subject: PAW 57 Forecasting Time Dimension Detection
I don't know what I've done different, but now have a 4 dimension cube working, but getting strange forecasts.
I have a product with historic sales between 0 (not sold in that season) and 3300 units yet two of the three forecasts are all forecasting negative sales, it appears to me that it's not auto detecting seasonality, so sees the decline in a winter product heading towards period 4 and expects it to continue into a negative number ???
------------------------------
Mark White
Original Message:
Sent: Mon October 12, 2020 07:17 AM
From: Mark White
Subject: PAW 57 Forecasting Time Dimension Detection
I now have this working, but only with a 3 dimensional cube as per the Tutorial, my dimensions are Product, Time and Version, if I add any additional dimensions e.g. Sales Measures (so I can select the measure I want to forecast e.g. Qty or Revenue) then it fails with the message
No element specified for at least one of the dimensions in the cube!
I have tried dragging the measure onto the rows (so it's not a filter in the header of the cube view) but still have the same problem, is it that Forecasting can only work on a 3 dimensional cube as I was hoping to forecast across multiple dimensions at the same time e.g. Product by Customer or Product by Region etc.
------------------------------
Mark White
Original Message:
Sent: Fri October 09, 2020 09:27 AM
From: STUART (MARTIN) PHILLIPS
Subject: PAW 57 Forecasting Time Dimension Detection
Hi Ryan,
The time dimension can have any name and can be made of more than one dimension.
The time series (that is made from one or more dimensions) need be continuous; and there can be issues when other reporting labels or multiple parent are in the time series too.
Sometimes creating a hierarchy with just the members to be considered in times series works.
Martin
------------------------------
STUART (MARTIN) PHILLIPS
Original Message:
Sent: Fri October 09, 2020 04:42 AM
From: Ryan Clapp
Subject: PAW 57 Forecasting Time Dimension Detection
They were, still not working.
Original Message:
Sent: 10/9/2020 4:41:00 AM
From: Florence Mougnaud
Subject: RE: PAW 57 Forecasting Time Dimension Detection
Hello Ryan,
Your "time(s)" dimension simply need to be placed in column when running the forecast or preview.
Regards,
https://www.ibm.com/support/knowledgecenter/SSD29G_2.0.0/com.ibm.swg.ba.cognos.tm1_prism_gs.2.0.0.doc/papr_forecast_features.html
------------------------------
Florence Mougnaud
Original Message:
Sent: Thu October 08, 2020 09:46 PM
From: Ryan Clapp
Subject: PAW 57 Forecasting Time Dimension Detection
What's the trick to get PAW 57 forecaster to recognize a time like dimension? Things I have tried
1) Alias's on/off
2) Cubes with 2 time like dim
3) Cubes with 1 time like dim
4) Typing into the selection boxes
Nothing seems to work. Is it expecting a dimension actually named "Time"?
Thanks!
------------------------------
Ryan Clapp
------------------------------
#PlanningAnalyticswithWatson