This site uses different types of cookies, including analytics and functional cookies (its own and from other sites). To change your cookie settings or find out more, click here. If you continue browsing our website, you accept these cookies.
I played around with the Assisted Modeling Tool. Here are my comments:
Value: I think it has a LOT of value for a novice data scientist or a "citizen data scientist" since it automates much of the heavy lifting of the modeling process. It is also a good tool to use to train a data scientist in the standard modeling methodology. It tries to keep every step of the process as simple and easy to understand as possible while also allowing a more experienced data scientist to make more complicated fine tunings within a couple of clicks.The embedded glossary is also appreciated and it will help novices quickly learn the DS Lingo. Also, in the "Select Algorithms" area, detailing the pros and cons of each of the algorithms is insightful as well.
I am assuming that the Assisted Modeling tool uses Python modeling libraries instead of R packages that the Predictive Tools use. The problem is that a saved model built with Assisted Modeling cannot be used with any of the Predictive Modeling tools like SCORE, MODEL COMPARISON< and LIFT CHART. In place of SCORE is a new tool called PREDICT VALUES. It would be nice if there were one version of PREDICT that would work for R generated models as well as Python generated models.
One of my favorite tools is the LIFT CHART. It would be nice to have the current LIFT CHART also work with Python generated models.
I am assuming that the PREDICT tool also incorporates all the TRANSFORMATION steps that were used in the Assisted Modeling process. So for example, if a variable had a specific imputation method and a had one hot encoding as well, these would then need to be applied to the data that is fed into the PREDICT tool before running through the model. Some additional transformations you should consider adding to the TRANSFORMATION tool are:
Transformations of a Numeric feature like 1/x, x^2, sqrt(x), log(x), e^x, etc. (use trending to determine which of these transformations is recommended for a numeric feature)
Multiplication of Numeric features like x*y, x*z, y*z, x*y*z
Addition of Numeric features like x+y, x+z, y+z, x+y+z
It would be nice to add in Neural Networks to the list of Algorithms attempted.
In addition I think it's disgraceful that Alteryx is asking for extra license fee to unlock these ML tools ! It's hard enough as it is to promote the tool in a large enterprise where there are competing tools and considerable resistance from IT folks. This is adding an insult to injury. Not a wise move Alteryx, sorry to say that !