hello,all:
In the process of one of my data mining projects using Microsoft SQL Server2005,i met a problem. I have created a mining model with microsoft_time_series algorithm and have processed it successfully.when i predict the pridictabe column [traff] with the following DMX:
select predict([traff],-50,0) from my_model_name
and:
select predict([traff],40) from my_model_name
(in this mining model,the values of microsoft_time_series algorithm parameters are : historic_model_gap=10,historic_model_count=10,and the other parameters i use the default values)
The problem is:the prediction results will contain some null values most of the time.And when i changed one of or some of the 9 algorithm parameters(minimum_supportcomplexity_penalty/historic_model_gap/ historic_model_count/......and so on),the position of null values will be different from before,even sometimes the prediction results will contain no null values. but i can not find any rules or ralations between the null values' existence and the 9 algorithm parameters.
This problem upsets me completely.Could somebody help me?
Time Series algorithm returns null when predictions become unstable. Where these nulls start depends not only on algorithm parameters, but also on input data the model was trained with. To understand how values of different algorithm parameters affect the model created (and predictions the model returns) you can look at the documentation. To understand these dependancies it is also helpful to understand how the algorithm works. Good starting point is following MSDN article:http://msdn2.microsoft.com/en-us/library/ms174923.aspx
Predictions usually are more stable when you increase values of COMPLEXITY_PENALTY and MINIMUM_SUPPORT parameters. These parameters inhibit tree growth which, provides more stable predictions.
Let us know if you have specific questions and we will try help you.
Tatyana
|||thank you very much for Tatyana's kind help.i will try again and check the result.
sql