Community Spring Cleaning week is here! Join your fellow Maveryx in digging through your old posts and marking comments on them as solved. Learn more here!
The Product Idea boards have gotten an update to better integrate them within our Product team's idea cycle! However this update does have a few unique behaviors, if you have any questions about them check out our FAQ.

Alteryx Designer Desktop Ideas

Share your Designer Desktop product ideas - we're listening!
Submitting an Idea?

Be sure to review our Idea Submission Guidelines for more information!

Submission Guidelines

Featured Ideas

0 Likes

It would be great if it was possible to output the top most influential features in producing the score for each individual entity/row when using the predictive and machine learning tools.

 

Similar to the way they work in DataRobot. Details here and here.

 

This would enable some simple interpretation of how a model came to an individual prediction and the most important features in that particular row/case.

0 Likes

Could you please support Redshift as well?

 

https://help.alteryx.com/2018.2/PredictiveAnalytics.htm

 
0 Likes

When I use the PCA tool, I run it with 2 PCs, then look at the results output to choose my principal components, and re-run it with the actual number of PCs that I need. I use the loadings and variance data quite a lot - it would be great to be able to output the loadings, variance, and also the scaled variables as data to work with in Alteryx, rather than just browse it in the report.

 

Gwilym_0-1588787962265.png

 

I have modified the PCA tool to do this myself, but I find I need to do this with each upgrade just in case anything has changed in the tool. I'd love it if the report summary data and scaled data was available as an output!

 

For reference, my amended version is available here:

https://gallery.alteryx.com/#!app/Principal-Components-Analysis-Extra/5eb2f79d0462d70bc0b6c516

0 Likes

Linear Regression Tool errors out with my data set if I sample more than 1 in every 31 cases. The sample size error-out is very consistent, despite the fact that different R error messages filter up with different runs. Support recommended small sample for the predictive tool then submit all data to Score. That's backwards of my need, which is to submit detail data to the predictive tool to create as precise a model as possible then apply that model to predicting a smaller set of future case outcomes. 

 

I used version 2019.4.8.22007. My full data set had 15.46 million rows, and one string field (which is necessary) accounts for the majority of predictors submitted to the model. I ran it from the Desktop version. The PC had 64 GB RAM and I even changed the default Virtual Memory settings in hopes that'd help.

0 Likes

R has a very large number of useful packages and examples.  Often, we only need a few lines of R code.  However, integrating that with the data flow in Alteryx can be complex.  It would be ideal if there was a tool where you could drop in R code, and have the tool create named inputs and outputs for each variable in the R code, and create blank text documents or YXDBs with the correct column names and variable types.  This seems like it could be automated, and would eliminate a lot of trial and error in using small pieces of R code for specialty tasks.

0 Likes

Since we know Alteryx uses R for a lot of its predictive and data analysis tools. It takes a while to run the workflow whenever there is R based tool is involved. I was told by a solution engineer that its because its opening and closing R in the background.

 

Sometimes my workflow has a bunch of tools which are running R in the background and it takes forever to run the workflow.

 

I think there should be a user setting which allows user to choose if the want to start R along with Alteryx and keep it running in the background.

 

Thanks,

0 Likes

Can you please enable option to select multiple performance measures in AB Trend tool. 

Currently we can only choose one metrics

0 Likes

@KuoL 

 

Yes, I know, it's weird to have a situation where a decision tree decides that no branches should be created, but it happened, and caused great confusion, panic, and delay among my students.

 

v1.1 of the Decision Tool does a hard-stop and outputs nothing when this happens, not even the succesfully-created model object while v1.0 of the stool still creates the model ("O") and the report ("R") ... just not the "I" (interactive report). Using the v1.0 version of the tool, I traced the problem down to this call:

 

dt = renderTree(the.model, tooltipParams = tooltipParams)

Where `renderTree` is part of the `AlteryxRviz` library.

 

I dug deeper and printed a traceback.

 

9: stop("dim(X) must have a positive length")
8: apply(prob, 1, max) at <tmp>#5
7: getConfidence(frame)
6: eval(expr, envir, enclos)
5: eval(substitute(list(...)), `_data`, parent.frame())
4: transform.data.frame(vertices, predicted = attr(fit, "ylevels")[frame$yval],
       support = frame$yval2[, "nodeprob"], confidence = getConfidence(frame),
       probs = getProb(frame), counts = getCount(frame))
3: transform(vertices, predicted = attr(fit, "ylevels")[frame$yval],
       support = frame$yval2[, "nodeprob"], confidence = getConfidence(frame),
       probs = getProb(frame), counts = getCount(frame))
2: getVertices(fit, colpal)
1: renderTree(the.model)

The problem is that `getConfidence` pulls `prob` from the `frame` given to it, and in the case of a model with no branches, `prob` is a list. And dim(<a list>) return null. Ergo explosion.

 

Toy dataset that triggers the error, sample from the Titanic Kaggle competition (in which my students are competing). Predict "Survived" by "Pclass".

0 Likes

The error message is:

 

Error: Cross Validation (58): Tool #4: Error in tab + laplace : non-numeric argument to binary operator

 

This is odd, because I see that there is special code that handles naive bayes models. Seems that the model$laplace parameter is _not_ null by the time it hits `update`. I'm not sure yet what line is triggering the error.

0 Likes

 

The CrossValidation tool in Alteryx requires that if a union of models is passed in, then all models to be compared must be induced on the same set of predictors. Why is that necessary -- isn't it only comparing prediction performance for the plots, but doing predictions separately? Tool runs fine when I remove that requirement. Theoretically, model performance can be compared using nested cross-validation to choose a set of predictors in a deeper level, and then to assess the model in an upper level. So I don't immediately see an argument for enforcing this requirement.

 

This is the code in question:

if (!areIdentical(mvars1, mvars2)){
        errorMsg <- paste("Models", modelNames[i] , "and", modelNames[i + 1],
                          "were created using different predictor variables.")
        stopMsg <- "Please ensure all models were created using the same predictors."
      }

As an aside, why does the CV tool still require Logistic Regression v1.0 instead of v1.1?

 

And please please please can we get the Model Comparison tool built in to Alteryx, and upgraded to accept v1.1 logistic regression and other things that don't pass `the.formula`. Essential for teaching predictive analytics using Alteryx.

 

0 Likes

Some of the predictive tools put out a "Score" field when output is run through the scoring tool, and some put out a "Score_1" and/or "Score_0".  Since I frequently reuse the same workflow template for different predictive model types, it would be nice if they were consistent so that I wouldn't have to crash the workflow the first time through to get the input field names correct for downstream tools (e.g., Sort).  Thank you

0 Likes

I have three groups: a control group, a group that got product A, and then a group that got product B. There is a way to test the differences across all groups rather than running separate t-tests (which introduces type I error several times). If my outcome is the percent of people who were contacted, I want to see if the percent is different across groups.

 

Control Group % who were contacted: 10%

Product A group % who were contacted: 25%

Product B group % who were contacted: 33%

 

I shouldn't have to run a t-test comparing control to A, then another comparing control to B, and then a third comparing A to B. I know the method is pairwise comparisons but I'm not finding how I can do this in alteryx and I've looked on the community and surprisingly the answer seems to be "you can't" but this is not a rare statistical test!

A product analyst at alteryx help build a macro in R to run the tests but the variables need to be categorical rather than continuous. The ideal solution is that an additional predictive analytics tool can run these ANOVA tests and there's something to specify whether the variable is categorical or continuous.

 

-Justin

0 Likes

Alteryx current C5 implementation requires very carefully crafted data to work.

This is fine in an academic environment, however production datasets are rarely available with such low level of tolerances.

 

Competing products have implementations that will not fail as easily and Alteryx doesn't have any other multi-way decision tree option.

A good alternative would be otherwise a tool that does some dataset diagnostic so that we at least can know where the problem is without having days of data scientist time to comb the dataset.

 

Thank you,

Marco

0 Likes

When you get an error message in an R tool it's almost impossible for a newbie to figure out what has just happened...

 

For eg. the normal R package randomForest sets 32 as a max number of classes for a given class variable

Thus when you happen to run randomForest on anything with > 32 classes you get an error, imposssible to figure out without searching on the net or better surfing on the community.

 

How about a basic rules checker providing message on the configs sayin;

 

"X" and "Y" variables are categoric and have more than 32 classes,

you have to fix thembefore running an RF tool in order to succeed"

 

0 Likes

I will be using TS Model Factory tool for running ARIMA quite extensivelty and understand that existing tool does not allow for model customization options. I will really appreciate if we can have customization options which helps to specify order of AR or MA components like it can be done in ARIMA tool.

 

Thanks!

0 Likes

There is a great functionality in Excel that lets users "seek" a value that makes whatever chain of formulas you might have work out to a given value. Here's what Microsoft explains about goal seek: https://support.office.com/en-us/article/Use-Goal-Seek-to-find-a-result-by-adjusting-an-input-value-...

 

My specific example was this:

 

In the excel (attached), all you have to do is click on the highlighted blue cell, select the “data” tab up top and then “What-if analysis” and finally “goal seek.” Then you set the dialogue box up to look like this:

 Set cell: G9

To Value: 330

By changing cell" J6

 

And hit “Okay.” Excel then iteratively finds the value for the cell J6 that makes the cell G9 equal 330. Can I build a module that will do the same thing? I’m figuring I wouldn’t have to do it iteratively, if I could build the right series of formulas/commands. You can see what I’m trying to accomplish in the formulas I’ve built in Excel, but essentially I’m trying to build a model that will tell me what the % Adjustment rate should be for the other groups when I’ve picked the first adjustment rate, and the others need to change proportionally to their contribution to the remaining volume.

 

There doesn't really seem to be a way to do this in Alteryx that I can see. I hate to think there is something that excel can do that Alteryx can't!

Top Liked Authors