Connecting a model with Excel for real time use
kylejohnson
MemberPosts:7
Newbie
Hello,
Beginner user here. I have searched all over for this but I assume I'm missing something. My situation is that I have created an Excel file that takes in streaming stock market data and makes a prediction about the short term movement of the market. I have created a model in RapidMiner that is very good at determining how useful a given prediction will be and I am now trying to figure out how to make this process work in real time. I have two ideas for how to make this happen and I'm wondering if there is anyway to make either idea work:
Idea 1 - Take the model's logic into Excel and run each example through the model within Excel. I have done this with simple decision trees by hand before but the RM model that I find to be very useful is a Gradient Boosted Tree model and obviously creating 400 trees by hand in Excel is way too much work. Is there a way to export the logic of the model in to Excel, either as a sheet or in VBA code?
Idea 2 - Each minute import a single example with its 20ish attributes, run it through the model and then pass the resulting prediction back to a specific cell in the same Excel file. I have dabbled with the Advanced Reporting Extension created by Old World Computing but I haven't gotten it to work the way I'm hoping.
If there are other methods that would accomplish the same thing, I would love to hear about them but I'm not knowledge enough yet to know of a third way. Thank you in advance for your time,
Kyle
Beginner user here. I have searched all over for this but I assume I'm missing something. My situation is that I have created an Excel file that takes in streaming stock market data and makes a prediction about the short term movement of the market. I have created a model in RapidMiner that is very good at determining how useful a given prediction will be and I am now trying to figure out how to make this process work in real time. I have two ideas for how to make this happen and I'm wondering if there is anyway to make either idea work:
Idea 1 - Take the model's logic into Excel and run each example through the model within Excel. I have done this with simple decision trees by hand before but the RM model that I find to be very useful is a Gradient Boosted Tree model and obviously creating 400 trees by hand in Excel is way too much work. Is there a way to export the logic of the model in to Excel, either as a sheet or in VBA code?
Idea 2 - Each minute import a single example with its 20ish attributes, run it through the model and then pass the resulting prediction back to a specific cell in the same Excel file. I have dabbled with the Advanced Reporting Extension created by Old World Computing but I haven't gotten it to work the way I'm hoping.
If there are other methods that would accomplish the same thing, I would love to hear about them but I'm not knowledge enough yet to know of a third way. Thank you in advance for your time,
Kyle
Tagged:
0
Best Answer
-
kayman
MemberPosts:662
Unicorn
For idea one I'm afraid the short answer will be no. Models in itself are basically complex equations using your example as variables, and this is not something you can easily translate to VBA (would be cool though...)
Why not trying to integrate the streaming stock logic directly in RM using one of the web extension operators? I assume your excel logic is making use of an API so this can be integrated in RM directly. This way you could kind of bypass the whole excel import complexity and still save the output to excel if needed.
7
Answers
the usual way to do this is to use a web service by RM Server. You can then write a query like
=WEBSERVICE("http://serverurl/:8080?value1=A1&value2=B2)
the result of this is an XML which you can use with FILTERXML to get the results. I do this frequently with customers to have a dirty but quick version of it.
I think either the idea@kaymanis proposing is the cleanest. The other way is a process which works like "Read Excel, Score, Write Excel" which is cron-jobed or triggered by you is the best version doing it.
Best,
Martin
Dortmund, Germany