How to save forecasting result after using series_decompose_forecast in Kusto?

Hi Vishal Vyas,

Thank you for posting query in Microsoft Q&A Platform.

The reason you are only seeing the first 7 data points in the time chart is because the series_decompose_forecast function is only forecasting one week forward, and your time series data spans 5 weeks.

To save the forecast results after using series_decompose_forecast, you can use the project-reorder operator to select the columns you want to keep and reorder them as needed. For example, you can modify your query as follows:

In this modified query, the project-reorder operator is used to select the Timestamp, y, and y_forecasted columns and reorder them as needed. The resulting forecast table can then be used to access the forecasted data points.

I hope this helps! Let me know if you have any further questions.

Hi @ShaikMaheer-MSFT

Thanks for you reply! This is merely a coincidence that I gave example of 5 week forecast in above issue. Even when I am forecasting monthly data of 10 months, I could see only first 7 data points.

Project-reorder only reorder the columns but I am interested in using the Time-stamp, y, y_forecasted data points in other function.

For example: In the code below, we are creating a time series of 22 months and generating a forecast 10 month forward. Here also extend y_forecasted shows first 7 data points. How to get all the data points which can be used in any other function

let ts=range t from 1 to 12 step 1 // generate 12 months of monthly data
| extend Timestamp = datetime(2023-01-01 00:00) + 30d * t
| extend y = 2*rand() + iff((t/24)%30>=5, 5.0, 15.0) - (((t%24)/10)*((t%24)/10)) + t/72.0 // generate a series with monthly seasonality and ongoing trend
| extend y=iff(t==3 or t==6 or t==9, y-8.0, y) // add some dip outliers
| extend y=iff(t==4 or t==7 or t==10, y+8.0, y) // add some spike outliers
| make-series y=max(y) on Timestamp from datetime(2023-01-01 00:00) to datetime(2024-10-01 00:00) step 30d; //create a time series of 22 months (last 10 months are empty)
ts
| extend y_forcasted = series_decompose_forecast(y, 12)  // forecast 10 month forward
| render timechart