I’ve been filling in the tools as I find them (slow process) but have run into a type of analysis that I can’t seem to find tools for in Knime. I’m sure it’s a nomenclature issue.
Analysis Process Overview
1. clean transaction data and filter to have the window of data you want
2. aggregate data so it describes performance in periods that make sense (weekly,monthly, etc)
3. link Unit data to transaction data
4. you might need to filter out some more data based on store parameters if there is something that eliminates them from being a control
5. use scatter plot and Plot of means to see if there is correlation in variable to use for Control selection
6. then use the AB Trends tool to get trend and seasonality data for each unit
7. use AB Controls tool to generate Control Units selecting numeric characteristics
8. Now you can analyze Lift with the AB Analysis Tool - connect the control Unit mapping from the AB Controls Tool and connect the Treatment Units (this is usually a standalone list) and then finally connect the Performance Data - map the unit fields and the performance field
From what I understand from the Alteryx website these modules use R functions beneath. Which is what often is the case with this software. So one approach could be to find an example of A/B testing in R and bring that to KNIME and R.
What is often very useful is to use KNIME to prepare and clean the data and to ‘hold everything together’ and use R (or Python) to do some special jobs.
Thanks so much for the suggestions. I come from using Alteryx and was hoping to do all the analysis in a GUI. My Python skills are best described as rookie.
Thanks again. So nice to see a active helpful community.
To an extend it seems to be the business model of some tools to provide a user interface to a script language like R. Which is perfectly fine. From my experience with some examples and even a limited understanding of coding you can come quite a long way and maybe understand more about the possibilities of certain tools and packages.
Concerning your analytical task it would be interesting to have data representing you challenge to see if this is possible to be done in ‘pure’ KNIME. I always found it difficult to think about such tasks in the abstract - since KNIME offers so many ways and nodes to handle tasks it is best to sketch out a way while working on data. Which is why I like working examples with data you can play with.
You could see a few examples of R and KNIME here. For most of them you would not have to be a super coder to understand and modify them - a lot are taken from web examples and adapted to be used in KNIME.
