format t tod2.; ... View more

Hi, Could you describe the problem in more detail? Maybe with a small example? Do you want to allocate marketing spends to different campaigns? Usually PROC OPTMODEL is the right tool to solve such problems. ... View more

Hi Bill, If you set 0 to 1 by 0.01 for each z parameter, NLIN first will do a grid search: will try each combination of the z starting values. This is 100**9 combinations. In this specific case I don't think it is necessary to specify such a fine grid for starting values. One starting value will be enough, because your model is linear, it will always converge. You could also convert (reparameterize) this problem and solve with proc reg or glm. You don't need to specify a formula for the derivatives with the DER statement. It is just a note. DER statement is needed in special cases, when you know the formula for derivatives better then SAS You can use the bound statement to restrict z parameters: BOUNDS z1-z9 <=1; ... View more

I never called a SAS program from SimStudio, but I would try the following: Use SAS Program Block, triggering SAS code execution at the end of the simulation day with the InSubmitCode port. Use InQueueData or InServerData ports together with Stats Collector Blocks or Dataset Writer Block to provide data from SimStudio to the SAS program. Use Dataset Holder Block to read data into SimStudio provided by the SAS program. ... View more

If you want to illustrate on a chart, that time is continuous through midnight then this is a (rare) case when a radar chart is adequate: data time;   t=1;   do i=1 to 10000;   t+1000*ranuni(123);   output;   end;   format t tod2.; run; proc gradar data=time;   chart t; run; If you want to identify "peak hours", "peak minutes" or something similar, then you are looking for the mode of the distribution with a "cyclical twist". I would follow what suggests: to re-center the data. data time2;   set time;   timePartOrig  =timepart(t);   timePartReCent=timepart(t+'12:00't); run; proc univariate data=time2 round=60 modes;   var timePartOrig timePartReCent;   histogram; run; You can see in this example, that I am using rounding (otherwise the is no mode of the data, because in my generated data time is continuous, and no value occurs twice.). If your data is similar, other procedures might be more useful to extract "peaks". (eg: prockde) ... View more

Hi, We only know that you have time series data of a bank about customers. Probably you have their monthly or daily balance or even their transactions (time stamped data). If you don't have SAS/ETS your can use STAT and of course Base. Some examples you can do with time series and time stamped data: -aggregating balances or transactions to hourly, daily, weekly levels overall, by customer or by some customer group. -aggregating by hour, by day of the week, day of the month - this way you can detect seasonality -you can use time as independent variable and balance or usage (or some other dependent variable) in a regression to detect trends -you can combine the two. For example create dummy variables from day of week, and use time as continuous variable -> detecting seasonality and trends -you can extract various aggregations from time series or time stamped data on various level, then use these as input variables (together with demographic and marketing varaiables) in a predictive model (to predict customer behaviour: balance, churn, propensity, etc.). Some technical stuff:   It is relatively easy to derive many complex aggregation measures for each customer (for example "number of ATM transactions two week before Christmas", "time difference between last 2 transactions") using a single data step.   proc means comes handy to calculate aggregations, I would emphasize here capabilities like extracting "the 3 most valuable transactions".   proc sql - as usual   proc rank - also useful to quickly calculate ranks (overall ranks, or ranks by customer) I would also suggest a search with the following search keywords: "time series datamining"   - since part of your problem (churn, direct marketing) is more related to this topic. Enterprise Miner has Time Series nodes: TS Correlation, TS Data Prep, TS Decomp, TS Dimension Reduction, TS Exponential Smoothing, TS Similarity. If you have EM, those nodes are specifically designed to handle these kind of data and problems. If you don't have EM, the procedures in SAS/STAT that do predictive modelling: REG, GLM, LOGISTIC, GENMOD, GLMSELECT (and many other). ... View more

data want;   set have;   flag=prxmatch('/.*(\d)\1{4,4}.*/',ssn); run; ... View more