Problem With Historical Prediction In Sales Forecast Model
Aug 2, 2006
Hi,
I have built a time series model to forecast sales value
I have data from jan 2004 to jan 2006 and the sales value is
at a day level in my database. But I am aggregating it to month level in the
DSV of the mining model.
I am required to make only historical predictions using the
above model starting form jan 2004 to jan 2006 for every month.
I have set Historical_Model_Count
and Historical_Model_Gap parameter
values to 24 and 10 respectively, and trying to predict for the past few months
(PredictTImeseries(SalesValue,-1,1))
But its throwing me the following error
Error(Data Mining): A time series
prediction was requested with a start time further in the past than the
internal models of the mining model, Sales Forecast, specified in the
HISTORIC_MODEL_GAP and HISTORIC_MODEL_COUNT parameters can process
In fact it throws the above error irrespective of what the Historical_Model_Count and Historical_Model_Gap parameter values
are
I am not able to figure our why this problem is happening?
What should the parameter values for the above scenario?
It would also be helpful if I can get an explanation on how
these two parameters affect the historical predictions. I kind of understand
that these two parameters are important for historical predictions but don€™t
know why or how.
I am new to SSAS and i want to try to build a "Sales" model. I will have some "Usage" data for some timespanns, but I am not quite sure how to tackle this. Is there somewhere a "Howto" for this?
Edit: There are several locations, and for each location a forecast is needed. And the Icing would be If I would be able to tell where my supplies must go 1st to achieve the best sales...
The potential Client wants to use Oracle but I would like to show them that SQL Server is the better tool for this ;)
I have few more clarification regarding time series. Firstly In my model the month level product sales value represented across 1st day of every month. So that the key time column is of datetime datatype containing a sequence of dates representing the 1st day of every month of the year. Eg: 2006-01-01, 2006-02-01€¦€¦€¦. etc. all in (yy-mm-dd) format But when I make prediction for next five months, though it makes monthly predictions the date part for the months are random whereas I expect the date part to be 1st of every month. What is the reason for this and how can I overcome it. Secondly Predicted sales values for some time period are negative though I do not have any negative value in the training data. What is the reason for this and how can I rectify it?
Thirdly In one of your earlier posts you had said that the time series algorithm does not have any built in time intelligence but uses the key time column as a time sequence stamp. So If have to make predictions for a particular time period where the time slice for each time period is 25 days or 50 days etc, then I understand that the input data used to train the model should be in the same time sequence. Or Can I specify the span of the time period according to which the prediction needs to be made? Basically how can I use the same time series model to make monthly, yearly, quarterly, daily or predictions or for custom time period like I have mentioned above.
I want to calculate the sum of actual sales until a date and forecast sales after a date.I am not sure what the best approach to this problem is, but I have tried my best with the following approach. Any better ways to solve this (using DAX).
I have created a parameter table that offers the last date of each month as possible choices to the user. I have tried to create a measure that sums actual sales up until this date.
I am new to Data mining in SQL Server. I am using SQL 2005 to create a forecast model for Product sales. I two fact tables that I am using. One has all the Orders historically with the line item details. The other table is a time dimension table which has the value of each of the time values referenced in the Orders. So the time dimesion has multiple values for one day as it goes down to the hour the order was placed.
Can I create a forecast using the time series algorithm in the Data mining module. When I tried to use this algorithm, I got an error that the time is not synchronized with starting series "Unknown" and I should try to set the missing_value_Substitution parameter to previous.
Can some one explain to me how this will resolve my issue. I read in one of the articles on Data mining that in order to use the time series algorithm , I need to have unique set of values for the time. Can some one help me with this
I have a market basket model using associations. It generated several dozen itemsets. However when I attempt to run a singleton prediction like this:
select (Predict(Orderproduct3q,INCLUDE_STATISTICS,10)) as [Recommendation]
From
[Case All]
NATURAL PREDICTION JOIN
(SELECT (SELECT '16407' AS [Pname])) AS t1
the resulting predictions don't take the itemsets into account. Instead, the predictions consist of the ranked products in the training set, ordered by frequency. This appears to happen regardless of the precise query specified within the "natural prediction join".
What's going on here and how do I generate a singleton prediction which makes use of the itemsets?
Is there a way to display the actual predicted value for an output attribute for a particular model. For example, say I am trying to predict if a particular customer is going to take advantage of a promotion (0=no, 1=yes) and I use neural networks. I know that I can use "Predict" to give me the prediction "yes" or "no" for each customer. However, the neural network actually spits out a number as a result. For example, a 0.997 would be interpreted as a "yes" for life insurance promotion. I do not want the probability that the prediction is correct. I want the actual output for the network.
The reason being is that I want to compute an error rate between the predicted value and the acutal value (root mean squared error or some other measure). Is there a way to compute this using the mining model prediction tab design view? I do not want to write the actual query as I teach a course in data mining using SQL Server and my students do not know DMX queries.
I have a question about writing a prediction query against a clustering model that has the same column added more than once.
Per Jamie, I can accomplish some crude weighting by adding a column to my model multiple times. See this post for an explnation... Now that I have that worked out, I was wondering how my DM query would look? If I have Input_A1, Input_A2 , & Input_A3 all being source from the same column in my structure do I have to reference all three when writing my prediction query?
HI Thanks a lot for your answer I say my request but I don€™t know why anybody don€™t answer me I have a project about predicting a value about selling and buying of a good like t-shirts and I use data mining for my project ,so I should use time series algorithm ,that €˜s mean I have previous data about t-shirts for 11 months ago and now I should say for 12th month ,how many t-shirts are sale? My tables saved on the excel file and it is problem, because how should I use this table for building model? After building a model and structure and predicting the value of 12th month for this store in the mining model then I use this query in model in bi: €œSELECT PredictTimeSeries(amount) From [Forecasting]€? This query showed a column and prediction value. After all of that now I should show this value in the application so I use c# language for building it, so I use a form in c# then I add a button to form that with clicking on this button, I can connect to my structure and then show process of connecting in the panel, then with clicking on the other button I can use this query (€œSELECT PredictTimeSeries(amount) From [Forecasting]€? ) and after using I can see prediction value in the textbox, that€˜s mean the value of 12th month show on textbox. Form has two buttons: one button is for connecting to mining structure and other button is for sending PredictTimeSeries query to structure, one textbox, which is for showing the predicting value for 12th month for selling t-shirts, one panel for viewing the lift chart. Also you say You are building an application that programmatically creates mining structure and a model, and then you want to train the model and display some results. And I don€™t know how can I train my model, I should use a special code for it? If your answer is say, please explain that and then say that codes are for training. Please if you can send c# code for Sporadic of stage, please send that, I need to this code; my request is emergency for me. Thanks a lot i am very sad because any body don't answer me
Read only transacted replication to about 60 clients. 100 or so transactions a day (8 hours), transaction (record) is probably about 8K/transaction. Reliable connection over 100base-T. Latency - 15 minutes would be great.
This doable? -- I couldn't find a timing/ sizing model any place.
I have a database table which has all the inputs, key and the result. In visual studio, I created a decision tree model which has exactly the same fields as in the table. However the visual studio automatically add space preceding the capital letters. As the field name in the Datamining model and those in the database table are slightly different. I cannot use NATURAL prediction join. Is there anyway to told the visual studio not to add the spaces in the variable names?
I am using BI Dev Studio for SS2005 in a research (as opposed to a production) environment. Often I want to compare the results of multiple models using the same attributes. If I switch to a different model, the Design view completely resets. Is there any way to retain the same field names with different models in the Design view?
My current workaround is to give my models similar names with AR, DT, CL, LOG, NN suffixes and make global changes in the DMX.
I have consulted the following without finding an answer: http://msdn2.microsoft.com/en-us/library/ms178445.aspx http://msdn2.microsoft.com/en-us/library/ms175642.aspx http://msdn2.microsoft.com/en-us/library/ms175678.aspx http://msdn2.microsoft.com/en-us/library/ms175637.aspx
I have built a Sales Forecast model to predict the sales value. Along with making historic predictions for previous time periods I also want to retrieve the actual sales values for those periods.
How can I achieve this in a time series model?
I also would like to know how do mining models store the data.
Do they store the data in the same table/view format as their respecive data source view or in the Model Content format.
I have a very simple time series model which processing works fine without any problem. However when I run the following query
SELECT
[TimeSeries].[PriceChange],
[TimeSeries].[Symbol],
PredictTimeSeries(PriceChange, -3, 2)
From
[TimeSeries]
WHERE
[TimeSeries].[Symbol] = 'x'
I get the following error:
TITLE: Microsoft SQL Server 2005 Analysis Services ------------------------------ Error (Data mining): A time series prediction was requested with a start time further in the past than the internal models of the mining model, TimeSeries, specified in the HISTORIC_MODEL_GAP and HISTORIC_MODEL_COUNT parameters can process.
The following is the excerpt of the minding model script related to the two parameters:
<AlgorithmParameters>
<AlgorithmParameter>
<Name>MISSING_VALUE_SUBSTITUTION</Name>
<Value xsi:type="xsdtring">Previous</Value>
</AlgorithmParameter>
<AlgorithmParameter>
<Name>HISTORIC_MODEL_GAP</Name>
<Value xsi:type="xsd:int">1</Value>
</AlgorithmParameter>
<AlgorithmParameter>
<Name>HISTORIC_MODEL_COUNT</Name>
<Value xsi:type="xsd:int">10</Value>
</AlgorithmParameter>
</AlgorithmParameters>
These HISTORIC_MODEL_GAP (1) and HISTORIC_MODEL_COUNT (10) should accommodate PredictTimeSeries(PriceChange, -3, 2). Could anyone shed some light on this?
Hi, I am not getting Mining Accuracy Chart and Min ing Model Prediction Plz tel me how to do.And how to use the filter input data used to generate the lift chart and select predictable mining model columns to show in the lift chart
i am trying to find a way to link an 'initial' Sale ID of a product to 'future' Sale IDs of products that will trace back to the original Sale ID.For example, if I call the original sale , 'Sale ID #123', how can i link future Sale ID's (child[ren]) and all future sales to the original Sale ID #123? Can I use a Surrogate Key or similar function?
Parent:Sale ID #123 Children:Â Sale ID # 456, Sale ID #789, Sale ID #.....
how I can link the original Sales ID (Parent) to Sale ID's (child[ren]) of future purchases currently existing and in the future going forward?
Hi After building a model in BI, I want to view the chart of model in mining model viewer, in the chart tab I can just see one prediction value that means for my model do prediction for some time slice and in prediction steps I can specify how many steps, I want to show this chart In mining model viewer tab we can see the chart of prediction also decision tree and the chart is for showing all of value prediction, and with choosing prediction steps we can specify that show just one value prediction or two or several values. But sometime I can see just one value in chart and sometime I can see several values in chart, This difference is for my data or no? And also for viewing historic prediction I should choice €œshow historic prediction€? and before that I should set Two parameters: Historic_ model _count and historic _model _count, But I can€™t see historic prediction (sometime this happens) Please help me.
Using the TimeSeries algorithm, how do I forecast more than one time period ahead? I read in you book on page 182 that PredictTimeSeries function can take a parameter for the number of time periods you want to predict. Fore example, SELECT PredictTimeSeries(Bread,5) tells the algorithm to predict the next 5 time periods. Can you tell me how to change that parameter using the graphic interfaces?
Attached is a file that generates a sample dataset from which I want to forecast a value for column 'Parcel Count 2014' from tomorrow (30/10/2014) till the end of December (31/12/2014). The forecast should be based on values from column 'Parcel Count 2013' for the same Month.
Background is that I am recreating charts in SSRS that were being created in Excel. The source data is residing in a SQL Server database. I'm having a hard time coming up with a SQL query to provide a 6 month forecast. I can get my data into a pivot (within a stored procedure) in the following format:
I have create a simple tabular rolling forecast report (with cumulative totals) from today (October) thru the next 12 months that looks like this. It smartly works no matter when the report is generated, by starting with this CurrentMonth and moving forward by using 1,2,3,4,etc. in the dateadd: =MonthName(datepart("m",dateadd("m",1,Now())))
The report sample (formatting lost in dropping it in here):
Code Block
Close Pct October November December January February
25.% $0 $0 $26,625 $0 $0
50.% $237,500 $110,000 $262,500 $0 $0
75.% $56,250 $0 $891,075 $0 $0
90.% $1,051,830 $0 $0 $0 $0
Monthly Total $1,345,580 $110,000 $1,180,200 $0 $0
Cumulative Total $1,345,580 $1,455,580 $2,635,780 $2,635,780 $2,635,780
It is working fine....there doesn't seem to be anything wrong with it (all numbers total correctly, etc.), but it is very unelegant.....and I know there must be a better way.
In the righthand most month (which would be September 2008) column, I have a formula that produces the amount (the Monthly Total amount is the same):
Is there any way to copy my Data of 2015 to the Planning/Forecasting Value of 2016?
My question is based on that we use INFOR ION BI right now and there we can just add an Button in our reports wich physically copies the value from one year to the next year based on some other rules in the cube.
Now I need to make this example work with SSAS and Excel PivotTables but I cant figure out how.
I have absolutely no clue where and how to accomplish it. Do I use Calculations, do I use Actions, do i make it in the Dataview, Cube or directly in Excel?
I have a report which totals sales by customer. Then table footer has a grand total of all customer sales. I would like to get a percent of each customer's sales against the total sales. How do I get the sum from the table footer to use in an individual customer row?
I need to develop a language specific dwh, meaning that descriptions of products are available from a SAP system in multiple languages. English is the most important language and that is the standard. But, there are also requirements of countries that wants productdescriptions in their language.Â
Productnr Productdesc Language 1       product    EN 1       produkt    DE
One option is to column the descriptions, but that is not very elegantly. I was thinking of using bridge tables to model this but you have to always select a language in a filter (I think)..
I'm thinking of a technical solution, such that when a user logs on, the language is determined and a view determines whether to pick a certain product table specific for a certain language. But then I don't have the opportunity to interchange the different language specific fields in a report (or in my case PowerPivot).
We have our Production server having database on which Few DTS packages execute every night. Most of them have Bulk Insert stored procedures running.
SO we have to set Recovery Model of the database to simple for that period of time, otherwise it will blow up our logs.
Is there any way we can set up log shipping between our production and standby server, but pause it for some time, set recovery model of primary db to simple, execute DTS Bulk Insert Jobs, Bring it Back to Full recovery Model AND finally bring back Log SHipping.
It it possible, if yes how can we achieve this.
If not what could be another DR solution in this scenario.
I need some advise on how to create a historical database.
What is the best way of doing this? For example, should I create a new row if a column is changed in a row and Time Stamp all record? What happens when I have child tables link to a Header table?
I have been looking on the NET for methods of creating a historical database, but I cant find any.
I have a question about historical table. I have a table in Sql Server that keeps the history of my clients for the last 3 months. This table has a PK : client_code and rep_date_id. I want to do a query that returns me a client and to extract the date when i inregistreted for the first time.
I think about a cursor, but i don't know how to use it. My table looks like this structure: - client_code -activity_code -country_code -district_code -... -rep_date_id_n
And I want to return --activity_code -country_code -district_code -... -start_date -current_date
A general data design question:We have data which changes every week. We had considered seperatinghistorical records and current records into two different tables withthe same columns, but thought it might be simpler to have them alltogether in one table and just add a WeekID int column to indicatewhich week it represents (and perhaps an isCurrent bit column to makequerying easier). We have a number of tables like this, holding weeklydata, and we'll have to query for historical data often, but only backthrough the last year -- we have historical data going back to 1998 orso which we'll rarely if ever look at.Is the all-in-one-table approach better or the seperation of currentand historical data? Will there be a performance hit to organizing datathis way? I don't think the extra columns will make querying too muchmore awkward, but is there anything I'm overlooking in this?Thanks.
From what I've read this is called 'slowly changing dimensions'. Bassically the system I'm working on needs to store the history of certain data so that at any time a user can look up an old project and view it exactly as is, even though the associated parts might have had certain changes over time. From what I can tell Type 2 ( current and historical records are stored in the same table) seems to be the most popular. Type 4 (current records in one table and historical records in a seperate history table) seems like it would also work but I've been unable to find any articles comparing the two. Does anybody have any info on the dis/advantages of one v.s. the other?
