No luck rendering any Dashboard Visualizations which require "Target" data item.

[FerdH4]

What kind of data item is a "Target"?  Maybe I should ask instead, what kind of data item is acceptable as a "Target"? 

I would suspect that it's a Measurable data item - but clearly, I could be wrong.

I can't get any of the Visualizations which require a "Target" value to render.  I've tried several different types including, for example, the Spiral and Driver Analysis.  I've used different data items including, for example, Mearsurable and Integer.

Thank you for sharing inputs.

[MFGF]

What kind of data item is a "Target"?  Maybe I should ask instead, what kind of data item is acceptable as a "Target"? 

I would suspect that it's a Measurable data item - but clearly, I could be wrong.

I can't get any of the Visualizations which require a "Target" value to render.  I've tried several different types including, for example, the Spiral and Driver Analysis.  I've used different data items including, for example, Mearsurable and Integer.

Thank you for sharing inputs.

Hi,

Both continuous (eg measure value) and categorical (non-numeric groups or categories) values are supported as a target. You need to have the right spread of drivers in your data set for the predictive model to be able to find anything relevant. You also need to be using "enriched" data, where Cognos Analytics has already identified the heuristics within the data. For an uploaded data set, this is done automatically, but if you are using an older Package (from Framework Manager), you will need to make sure the package has been enriched.

Cheers!

MF.

[cognostechie]

MF - Is this supposed to be an alternate method of how a human being who is an expert in data analysis, is familiar with the database and is technically competent to decide the relationship within the data to determine and develop the complete analytical solution? Ex: Know what is valuable info for the organization, decide and develop how the dashboard should look like, what data elements it should have, what should be the target (child data elements, drill downs etc.).

[MFGF]

MF - Is this supposed to be an alternate method of how a human being who is an expert in data analysis, is familiar with the database and is technically competent to decide the relationship within the data to determine and develop the complete analytical solution? Ex: Know what is valuable info for the organization, decide and develop how the dashboard should look like, what data elements it should have, what should be the target (child data elements, drill downs etc.).

Hi,

Are you asking about package enrichment? Sorry, I'm still on my first cup of tea of the day - lol. Package enrichment is an option from the ellipsis when you click on a package in the portal.

The target being asked about here is for specific visualizations that call ML algorithms (eg Spiral, Driver Analysis) - for example, a flag in the data that marks whether a customer churned. The algorithm works out the most significant driver values from the other items in the data set, and presents them to the user in the visualization - so you see a visualization that shows you what drives customer churn.

Cheers!

MF.

[cognostechie]

You did answer it MF even without completing your first cup of tea  :) Yes, I was asking whether the predictive model and the enrichness of the package do what a human can manually do. Like we can manually determine and define when a customer churned by creating a calculation which would determine if there is a certain time gap between the current date and the last order date of the customer and comparing it to the average time gap between that customer's previous orders or based on some other factor whatever applies in this case.

It seems that all these features like ML, AI etc. are based on certain rules and applying those rules to look at the past data and predicting the future based on the past. I was an ERP programmer originally so was writing heavy transactional codes so am familiar with what we used to call 'rules engine' where all scenarios were defined. Somehow I see all these new technologies as the same as rules engine, just named differently. It does what a programmer would do so it definitely replaces the programming effort and reduces the effort on the report developer so it's good that way.

[MFGF]

You did answer it MF even without completing your first cup of tea  :) Yes, I was asking whether the predictive model and the enrichness of the package do what a human can manually do. Like we can manually determine and define when a customer churned by creating a calculation which would determine if there is a certain time gap between the current date and the last order date of the customer and comparing it to the average time gap between that customer's previous orders or based on some other factor whatever applies in this case.

It seems that all these features like ML, AI etc. are based on certain rules and applying those rules to look at the past data and predicting the future based on the past. I was an ERP programmer originally so was writing heavy transactional codes so am familiar with what we used to call 'rules engine' where all scenarios were defined. Somehow I see all these new technologies as the same as rules engine, just named differently. It does what a programmer would do so it definitely replaces the programming effort and reduces the effort on the report developer so it's good that way.

I think the benefit here is that it considers *all* the drivers it is given, and uses statistical algorithms to work out the strongest correlating factors that drive the desired target. If a human is doing this, they generally use experience and intuition to bring in what they *think* are the most likely data items. The danger is they may inadvertently miss something important - simply because they didn't think to bring that item into their analysis. The other issue it avoids is confirmation bias - an analyst thinks they know what is driving whatever they are investigating, so they bring in items that support their supposition - and again potentially miss things that may be far more important and relevant to the outcome. Rules engines are great for some things, but they look for a fixed set of scenarios. ML models are not constrained like that - and use statistics to drive the answers they produce across all items of data they are fed. Of course, the models need to be trained in order to be able to deliver accurate results. The new generative AI solutions now available make model training so much easier, and I'm sure we'll see a lot more of this in the coming months and years.

That's my Friday pontification over for this week :)

Cheers!

MF.

[FerdH4]

WAHOO!  I've never gotten five replies before on Cognoise.  I feel "enriched" today.  Thank you both cognostechie and MFGF.

I think the answers to the questions posed are:
1.  I've definitely got a few Measurable data items in my data package - but not nearly as many as I'd like!
2.  And, this data package has been enriched.  I've used it to sem-successful poke around in Exploration many times.

[cognostechie]

If a human is doing this, they generally use experience and intuition to bring in what they *think* are the most likely data items. The danger is they may inadvertently miss something important - simply because they didn't think to bring that item into their analysis. The other issue it avoids is confirmation bias - an analyst thinks they know what is driving whatever they are investigating, so they bring in items that support their supposition - and again potentially miss things that may be far more important and relevant to the outcome.

That's very true and the intelligence level of the person also becomes a driving factor in this. Most people have average or less intelligence so are likely to miss relevant things.

Have a great weekend !