Before implementing Qlik - R integration, In my organization i need to get clarified few of my doubts. I am posting few questions, to get clear idea. I am a manager with exposure to Qlikview , not R.
Flow of the data will be like Database --> Qlikview (Real time) --> R (modeling ) --> Qlikview .
1) What is the cost : servers, licences, connectors, resources, hardware ... 2) Can it be instantly predictive ? 3) Is it used anywhere for commesial modeling ? 4) I want to have auto-made models, as well as designed-by-me models, as well as imporrted models. Is it possible ? 5) I want to version my models, and version my sets of data. We could say that I want to apply any model on any data set 6) What will happen when my model will change (every 6 month about). Do I have to redesign the data set ? 7) If i redesign the data set, will i be able to run a new model on a previous data sets. Yes or no, depending on what ? 😎 Can I tag some data sets, subsets, ...Can I add some comments , share my data sets with my colleagues for trainig or testing purposes ? 9) If I want to use SPSS with Qlikview , is it advisable ?
10) Do I need different data sets for different predictions (early claims, fraudulent claims ....) 11) Can i get some dashboard to measure the efficiency of my models (before / after) 12) What is the limit of the system : number of variables, size of the data, ....
Is this integration process possible on enterprise level ? I am having very large and dyanamic datasets which are updated in real time . Is there is limit of this system like in terms of number of variables and size of data?
Database --> Alteryx workflows for data blending and R based predictions/estimations/clustering --> Alteryx based scoring (scheduled to be updated/rerun in a few moments or for each trigger) --> Discover your data in Qlik (Real time)
1) What is the cost : servers, licenses, connectors, resources, hardware...
50K USD Server or if it's a small deployment 5K USD for scheduling, 5K USD per designer licenses...
2) Can it be instantly predictive?
It may learn near-real-time with new performance data being added...
it can definitely score real-time
3) Is it used anywhere for commercial modeling?
Alteryx is used by several well known financial institutions, banks, insurance firms, management consultancies
Pure R may be used as well... But you'll need exceptional experience with that..
4) I want to have auto-made models, as well as designed-by-me models, as well as imported models. Is it possible ?
Data preparation and target definitions should be made by you
Several modeling techniques are almost like automatic and error prone(like random forest)
5) I want to version my models, and version my sets of data. We could say that I want to apply any model on any data set.
"Applying any model on any data set" is a little bit too vague and does not sound so wise...
6) What will happen when my model will change (every 6 month about). Do I have to redesign the data set ?
May not need to redesign the data set... especially if hypothesis, observation and performance periods are not changed.
Calibrating models will fix the weightings accordingly, you can calibrate more frequently and it's probably is better to do so...
7) If i redesign the data set, will i be able to run a new model on a previous data sets. Yes or no, depending on what?
😎 Can I tag some data sets, subsets, ...Can I add some comments , share my data sets with my colleagues for training or testing purposes?
Yes in alteryx and some in SAS and SPSS
9) If I want to use SPSS with Qlikview, is it advisable?
Possible
10) Do I need different data sets for different predictions (early claims, fraudulent claims ....)
You may need a data set with a totally different attribute set
Should design it according to your business
11) Can i get some dashboard to measure the efficiency of my models (before / after)
Tracking the current gini and trend of predictive power for classification models would be wise
RMSE or similar for regression
CCC for clustering etc...
12) What is the limit of the system : number of variables, size of the data, ....
For the row size, 20K rows is something like a statistical limit. may not matter much if you add more...
For the number of variables, finding more of orthogonal (uncorrelated in between) variables is always a good thing... but most models will use a small number of very effective/valuable key attributes... top 50 variable will most likely do...
